AI components

Shared-stack-first browsing for ai layers used across home and humanoid robots.

1002 Sensor 335 Connectivity 352 AI 70 Voice Assistant
Components glossary 30d trends All components

AI workbench

Quick orientation across all four component layers. The current layer is highlighted.

Sensor

Scan the perception stack first: mapping, vision, proximity, touch, and orientation.

1002

Shared

100

One-off

902

Top adoption

IMU · 40 robots

Browse layer 30d trends

Connectivity

See which radios, apps, and protocols repeat across robot ecosystems.

335

Shared

53

One-off

282

Top adoption

Wi-Fi · 115 robots

Browse layer 30d trends

AI

Current

Compare autonomy stacks, compute platforms, navigation brains, and branded intelligence layers.

352

Shared

2

One-off

350

Top adoption

Not Officially Disclosed · 2 robots

Browse layer 30d trends

Voice Assistant

Browse speech interfaces, assistant integrations, and voice-control patterns without the fluff.

70

Shared

11

One-off

59

Top adoption

Amazon Alexa · 33 robots

Browse layer 30d trends

AI directory

Shared components stay in the main scan path; one-off entries stay bucketed until you actually need them.

Directory layer

Shared stack first, long tail on demand

Use the repeated ai signals to narrow the field quickly, then open the single-use entries only when an exact vendor label matters.

Tracked

352

Shared

2

One-off

350

30d active

232

Shared leaders

What repeats across robots

Not Officially Disclosed
2 robots
Reactive AI Obstacle Avoidance (200+ object types); SmartPlan 3.0
2 robots
10 TOPS AI platform with LiDAR and vision fusion for automatic mapping, path planning, 300+ obstacle recognition, and smart cliff protection
1 robots

Fresh 30-day verification

What was touched recently

Reactive AI Obstacle Avoidance (200+ object types); SmartPlan 3.0
2 active
10 TOPS AI platform with LiDAR and vision fusion for automatic mapping, path planning, 300+ obstacle recognition, and smart cliff protection
1 active
360° LiDAR and dual-camera AI vision for mapping, path planning, and 300+ obstacle detection
1 active

Browse lens

How to read ai

This catalog mixes model names, compute platforms, autonomy stacks, and branded systems. The shared table surfaces reusable patterns; the long tail captures one-off marketing or deployment labels.

Open 30d trends

Shared stack first

Multi-robot components worth scanning first

These are the reusable pieces that recur across multiple robots, so they do the heavy lifting for fast comparison before you dive into the edge cases.

2 entries

Not Officially Disclosed

FX Aegis · Mornine M1

2
Reactive AI Obstacle Avoidance (200+ object types); SmartPlan 3.0

Qrevo Curv 2 Flow · Qrevo Edge 2 Pro

2

Single-use index

Collapsed one-off implementations

Keep the rare branded edge cases available without forcing the main browse path to slog through one-off shells row after row.

350 single-use entries

A-D

85 entries

Single-robot components kept off the main scan path
Addverb describes the ELIXIS platform as Physical AI-ready, with multimodal sensor fusion, 3D SLAM/navigation, reinforcement learning, imitation learning, model predictive control, and VLA-enabled planning; the company has not broken out every ELIXIS-W software specification publicly. Affective AI engine with long-term memory, multimodal sensing (voice + touch + interaction habits), and continuous personalization AGIBOT describes the D2 Max as AGI-driven with Level 3 autonomous operation; detailed compute hardware and model stack have not been officially disclosed. AgileCore platform; Google DeepMind Gemini Robotics integration (announced) Agility Arc Planning System AI computer vision for obstacle detection (people, pets, walls, curbs) on M20i model; NetRTK autonomous navigation on all models AI Dirtsense Monitors Floor Dirtiness In Real-time And Adjusts Cleaning Intensity Automatically AI motion-control gait algorithm with autonomous mapping, localization, path planning, and intelligent obstacle avoidance. AI obstacle avoidance, smart room mapping, autonomous scheduling AI recognition distinguishes people and pets, filters irrelevant footage, and organizes event recordings; no general-purpose onboard voice assistant has been officially disclosed. AI SmartSight obstacle recognition (240+ objects), auto room-type detection for MopSwap pad selection, pet and pet-waste detection (99.9% claimed) AI vision debris detection with first-run 3D mapping, smart route planning, targeted spot cleaning, and real-time positioning AI-based autonomous driving system with real-time obstacle detection and path planning (Pro) AI-based control, planning, and estimation for collaborative lifting, navigation, and intention-aware interaction AI-based motion control with multimodal sensor fusion, 3D spatial intelligence, and Hexagon mission control system AI-driven cooking-control system for programmable culinary workflows; underlying AI/computer-vision details are not publicly disclosed. AI-driven obstacle avoidance with 360° scanning, object recognition AI-driven whole-body control and end-to-end autonomy for industrial tasks; Noble says the robot platform can learn real-world skills through language-based instructions, demonstrations, and gestures, with sim-to-real work supported by Schaeffler actuator models and NVIDIA Newton physics simulation. AI-enabled flight logic for autonomous delivery-flight monitoring and management; detailed autonomy stack not publicly disclosed. AI-enhanced obstacle recognition using dual-laser 3D structured light and an AI RGB camera; Dreame claims recognition for more than 280 object types. AI-enhanced OmniSight navigation with upgraded binocular vision, proactive light, and recognition for 280+ object types AI-Inverter™ dynamic power adjustment, intelligent path optimization, infrared + IMU pool mapping AI-Perception navigation with 12 sensor types, sensor-fusion obstacle handling, intelligent dirt detection, and adaptive rewashing when an area still reads dirty AI-powered camera with green-spectrum LED illumination identifies nearly 200 hidden household substances and stains, adapts cleaning in real time, and checks between passes before moving on AI-powered customer interaction and drink recommendations, proprietary vision-AI pour control, NVIDIA Jetson Thor acceleration, and Azure AI enhancements for vision, voice, autonomous reasoning, contextual awareness, and operational alerts AI-powered detection, deterrence, and reporting stack for autonomous perimeter security patrols AI-supported navigation with recognition of 300+ object types, Smart Resume cleaning continuation, and automatic narrow-gap side-brush extension AI-supported pool recognition, 3D path planning, targeted turbo suction, and scheduled maintenance timers AI.See obstacle avoidance with RGB visual recognition for 200+ object types, combined with iPath laser navigation AI.See visual obstacle avoidance with RGB recognition, iPath laser navigation, and dirty-spot/stain detection for targeted mopping AIVI 3D 3.0 with VLM deep learning for object recognition; AI Instant Re-Mop for stubborn stain detection AIVI 3D 4.0 camera-based obstacle avoidance with object recognition and nighttime illumination AIVI 3D 4.0 obstacle handling plus app-marked FocusJet spot pre-treatment zones for dried stains AIVI 3D 4.0 Omni-Approach Technology with VLM deep learning neural networks for object recognition and semantic obstacle classification; AI Instant Re-Mop 2.0 for stain detection and targeted deep cleaning AIVI 3D 4.0 with enhanced Semantic Model for dynamic edge-distance adjustment; AGENT YIKO 2.0 autonomous AI agent for multi-step cleaning planning AllSense 3D Fusion (LiDAR + AI Vision), 10 TOPS chip, real-time 3D mapping with 210K+ point clouds/sec Anybotics Autonomous Inspection Stack For Navigation And Data Collection AONavi 2.0 navigation with RTK + VSLAM 2.0, 10 TOPS onboard computing, and OmniSight full-scene obstacle and terrain recognition Apptronik AI Platform ASUS Maestro AI agentic orchestration for personalized conversation, task coordination, and memory-based support ASUS Maestro AI orchestration with emotion-aware interaction, multilingual support, guided navigation, follow-me assistance, and domain-specific service workflows; detailed autonomy stack not officially disclosed. AuraVue 3D LiDAR-Vision Fusion; SmartPath AI for systematic path planning; Patch Free adaptive cutting power Automated recipe execution and cooking-control system integrated with SyncKitchen; underlying AI/computer-vision details are not publicly disclosed. Automated steam-cooking control for frozen noodle recipes; underlying AI/computer-vision details are not publicly disclosed. Autonomous Home Navigation And Voice-triggered Interaction Autonomous hospital navigation plus VSee AI Workflow Engine integration for clinical AI modules such as triage, documentation, alerts, and workflow automation; exact onboard compute stack not publicly disclosed. Autonomous hospital navigation with facility mapping/simulation, computer-vision navigation through elevators and doors, obstacle avoidance, and transport analytics; exact onboard AI stack not publicly disclosed. Autonomous locomotion, perception via stereo/laser/IR point clouds Autonomous narrow-path service navigation, real-time multi-robot path coordination, and stable delivery behavior; detailed autonomy stack not publicly disclosed. Autonomous navigation and inspection in darkness/extreme environments Autonomous ordering-to-serving workflow with Elite Robots cobot motion control, order management, smart auto-cleaning, auto fault recovery, sales reporting, and remote stock monitoring; no dedicated AI model details publicly disclosed. Autonomous stack with 3D environment mapping, object recognition, full-body motion planning/control, and task execution management Autonomous Task Execution With Periodic Status Checks Behavior-learning Interaction Model Tuned For Therapeutic Companionship Boston Dynamics AI Platform Boston Dynamics autonomy stack (autonomous navigation, dynamic replanning) Boston Dynamics vision and planning system (real-time decision making) BrainOS commercial autonomy platform (Brain Corp) Built-in Chess AI With 19 Difficulty Levels Plus Conversational Coaching And Analysis Built-in high-performance motion control algorithms, fully open SDK and hardware interface, and reinforcement-learning / embodied-intelligence research workflow support Built-in large language model with long-term memory and multimodal emotional perception across visual, auditory, and tactile inputs Carbon — General Purpose AI ChatGPT-based multilingual dialogue for kid-focused conversation, storytelling, games, and voice interaction ChatRomi 1.0 plus Lacatan-only ChatRomi 2.0; official guidebook says ChatRomi 2.0 uses OpenAI Realtime API, web search, optional visual understanding, customizable speaking styles, and long-term memory support after the April 21, 2026 Lacatan update CleanMind AI with 3D MatrixEye 2.0 — recognizes 200+ obstacles and 40+ stain types, adaptive cleaning by room and mess level Cognitive AI platform with multimodal reasoning, autonomous navigation (AMR), self-learning, predictive analytics, ERP/MES integration Cognitive AI platform with reinforcement learning, autonomous learning from environment interaction, NVIDIA partnership for sim-to-real transfer Computer vision-based clutter detection and home navigation with privacy-focused local processing according to Clutterbot's FAQ Control architecture not officially disclosed; WIRED reports launch footage showed GD01 operating without an onboard pilot during part of the demo. COSA (Cognitive OS of Agents) — physical-world-native agentic OS COSA (Cognitive OS of Agents) + VideoGenMotion (VGM) video-to-motion framework CraftNet hierarchical vision-tactile-language-action model for fine manipulation, pairing an interaction layer for last-millimeter contact with a motion-control layer CyberNav fusion navigation combines RTK, VSLAM, IMU, and wheel tracking; QuadVision AI obstacle avoidance recognizes 200+ object types Deep learning AI for natural conversation, face recognition, voice recognition, and adaptive learning Deep reinforcement-learning table-tennis control trained in simulation, with low-latency event-based perception, 31.25 Hz policy updates, and 1 kHz trajectory execution on the robot hardware Designed as a machine-learning-friendly humanoid research platform; WIRobotics says it partnered with RLWRLD for physical-AI development, but exact onboard compute and model details are not officially disclosed. Designed for embodied-AI data collection and model-inference workflows; official copy describes end-cloud collaborative data collection with automated validation and manual review, but does not disclose onboard compute hardware. Developer SDK, URDF/simulation support, whole-body policy training workflow, and optional NVIDIA Jetson Orin NX compute for local vision models, motion policies, and robot applications; public model details are not disclosed. DFAI (Design for AI) architecture — software-hardware integrated system for embodied intelligence Diligent proprietary AI stack with deployment-trained models; Moxi 2.0 powered by NVIDIA IGX Thor with 10x compute increase over Moxi 1.0, robot foundation model for dense navigation and complex manipulation Donut Robotics says Cinnamon Mini is built around video-based motion learning rather than traditional motion-capture workflows; model, compute, autonomy, and perception specifications are not officially disclosed. DoorDash Labs autonomy stack — deep learning + search-based path planning, real-time obstacle detection and avoidance Dreame describes APEX as using advanced AI for embodied yard maintenance; model architecture and perception stack have not been publicly detailed. Drone-derived Obstacle Sensing And Path Planning With Machine-learning Perception Dynamic symmetry / dynamic isotropy design framework with simulation-derived locomotion and control experiments; exact onboard compute and autonomy stack have not been officially disclosed.
E-H

47 entries

Single-robot components kept off the main scan path
Edge AI — compact multimodal model (vision + audio) with behavior engine trained on narrative vignettes; all processing runs on-device Edge computing up to 2070 TOPS; UniX AI embodied intelligence stack (UniFlex imitation learning, UniTouch tactile perception, UniCortex task planning) Educational rule-based behavior stack with ultrasonic obstacle sensing, rhythm-following microphone interaction, balance/stability behavior, preset action programs, and MakeCode/Python programmability; no LLM or cloud assistant is officially disclosed. EEVE Higgs local adaptive AI for train-by-demonstration tasks; camera-frame input to motor-control output; local training and execution on Willow X using EEVE eOS and an NVIDIA Orion processor EFLS 2.0 + Visual SLAM + VisionFence AI obstacle avoidance EFLS 3.0 positioning + VisionFence obstacle avoidance EFLS LiDAR+ triple fusion (LiDAR + NRTK + Vision), 200+ object detection, obstacle avoidance as small as 1 cm Efls Nrtk Positioning With Panoramic Visionfence Obstacle Avoidance And Geosketch Real-scene Mapping Embodied AI service stack with autonomous environment recognition, task understanding, precise task execution, VSLAM plus LiDAR SLAM mapping, 3D mapping, and intelligent obstacle avoidance. Embodied AI stack combining data-driven embodied intelligence, advanced AI model strategies, hierarchical high-level/low-level planning, generalized learning operations, and PuduFM/PuduAgent platform integration. Embodied AI stack with multimodal perception, semantic navigation, and reinforcement-learning-based task planning Embodied AI system powered by NVIDIA Jetson AGX Orin, Axonex AX-CORE database, AI vision recognition, grasping algorithms, LLM-based conversational interaction, programmable task library, and scheduled task execution. Embodied-AI development platform with ROS 2, Isaac Sim, MuJoCo, LeRobot, VR teleoperation, and data-collection compatibility; VLAI has not disclosed onboard compute or model details. Embodied-intelligence platform with whole-home mapping, visual recognition and obstacle avoidance, posture/motion tracking, multilingual conversational interaction, and support for open programming, VR integration, and reinforcement-learning tools End-to-end neural network motion control; education edition lists NVIDIA Jetson Orin NX (16G) End-to-end neural networks on NVIDIA Jetson Orin NX; trained via NVIDIA Omniverse, Isaac Sim, and Cosmos synthetic data pipelines Epos Satellite Navigation With Virtual Transport Paths And Geofence Controls ERA-42 end-to-end AI model — proprietary foundation model for embodied intelligence Face detection, object detection/recognition, skeleton-based imitation, speech recognition (STT), text-to-speech (TTS) Factory firmware includes an AI Agent for voice-based interaction, expressive animations, OTA updates, app-linked remote video/avatar control, and open development through Arduino, UiFlow2, PlatformIO, and ESP-IDF on the ESP32-S3 CoreS3 controller. Foundation Cortex Physics-informed AI Stack Connected To Phantom-mk1 Fourier AI Platform Fourier Full-Perception Multimodal Interaction System (dual-path: rule-based + LLM reasoning) Full-stack self-developed WALL-A large operating model with tens-of-billions parameter scale for perception, reasoning, and precision manipulation FullDepth says the robot combines fish neuro-drive models, AI algorithms, and high-precision attitude control to generate lifelike arowana swimming in automatic or remote-control modes. GAC in-house embodied AI stack with pure-vision autonomous navigation, localization, and autonomous decision-making Geekplus Brain embodied intelligence for warehouse picking, packing, box handling, inspection, and coordination with broader Geekplus automation workflows. Gemini Flash + ChatGPT-class models; MBTI-based personality modeling; long-term emotional memory Generative AI + pet behavior analysis, sound-based emotion recognition, AI pet recognition, intelligent tracking, V-SLAM navigation, and AIVI object recognition Generative AI + proprietary emotional AI models (Andromeda OS); face recognition, emotion detection, personalized memory, mood-adaptive responses Generative Bionics Physical AI platform for embodied control, with Fincantieri describing AI plus manipulation, perception, vision, and locomotion capabilities for welding-support validation. GigaAI positions Maker H01 as an AI-native physical-AGI body that can reason over flexible-object manipulation and decompose vague instructions into long-horizon action plans; exact onboard compute, model stack, cloud dependency, and developer interface details have not been publicly disclosed for H01. Google Gemini + proprietary Samsung language models GPT-4o mini integration + Visual SLAM navigation GPT-powered conversations; current official hardware section lists a 5 TOPS processor, Cortex-M4 100 MHz co-processor, dual-core DSP 360 MHz HIFI audio engine, 2 GB LPDDR4 RAM, and 8 GB eMMC 5.0 storage GPU (1,024 cores) + 32 Tensor cores + 8 CPU cores, 512GB storage (LOVOT 3.0) Haier embodied-home AI system integrated with AI Eye 2.0 appliance vision and the UHomeOS smart-home platform for household scene understanding, task coordination, and appliance collaboration Helix VLA (in-house vision-language-action model) Honda Distributed Control System Honda proprietary 3D processor (stacked dies: processor, signal converter, memory) Huawei Pangu Embodied Large Model; KaihongOS (OpenHarmony-based); external LLM ecosystem support Hugbibi describes mood/state interpretation, touch/sound/motion reactions, app-visible memories, and an MBTI for Pets personality system that changes reactions over time; exact onboard compute, cloud dependency, and model providers are not officially disclosed. Human-aware Autonomous Navigation And Whole-body Control Algorithms Developed By Aeolus Human-in-the-loop Teleoperation With Whole-body Coordination And Balance Control Humanoid's KinetIQ four-layer AI stack with end-to-end reasoning and skills powered by NVIDIA processing Hybrid autonomy combining Devanthro physical AI for chores/monitoring with human-in-the-loop VR teleoperation; compute platform uses Nvidia Jetson Orin + Orin Nano HybridSense AI Vision with 40+ debris-type recognition, adaptive path planning, real-time obstacle avoidance, and seven smart cleaning modes
I-L

37 entries

Single-robot components kept off the main scan path
Independent computing platform for autonomous route planning, 3D stair-data interpretation, and proactive connection with compatible X60 Pro Series robot vacuums. INFFNI/DOBOT describes intelligent subject tracking, follow mode, dual-vision smart perception, and terrain-adaptive autonomous mobility; exact autonomy stack not officially disclosed IntBot describes a proprietary multimodal social-intelligence engine that fuses vision, audio, and language in real time to coordinate speech, facial expression, and gesture, interpret social cues, understand intent, and adapt behavior in public settings. Intel Atom E3845 quad-core CPU, NAOqi OS (Linux-based) Intel Core i5-1135G7 + Jetson Xavier NX ×3 Intel Core i5/i7 + optional Jetson Orin NX (up to 3 compute units) Intel Core i7-1370P (14 cores); NVIDIA Jetson AGX Orin 32GB (200 TOPS); optional Edge LLM (MiniCPM) Intel i7 (real-time control) + NVIDIA Xavier (AI inference) Intel NUC i3 onboard compute with Intel Core i3 dual-core CPU, 8 GB DDR4 2666 MHz RAM, and 250 GB M.2 SSD; current ROBOTIS e-manual frames the 2025 re-release around ROS 2 + DYNAMIXEL SDK development Intelligent path planning with dynamic Z/N route matching, edge detection, obstacle-aware rerouting, and multi-mode cleaning control IONOSPHERE robotics AI, control, server, and software stack for industrial safety and European data sovereignty iPath 2.0 smart navigation with LDS+ laser mapping and obstacle avoidance iRobot OS with ClearView LiDAR navigation, obstacle avoidance, and Carpet Detect iRobot OS with Dirt Detective, PrecisionVision Navigation, AI obstacle recognition Irobot Os With Enhanced Dirt Detect And Dirt Detective Room-priority Intelligence Irobot Os With Object Avoidance And Room-level Cleaning Automation Irobot Os With Precisionvision AI Object Recognition And Dirt Detect Mess Prioritization KEENON describes XMAN-F1's WAIC demonstrations as using multimodal interaction and large language model technologies; implementation details are not publicly disclosed. Kid-focused conversational AI with moderated, age-appropriate interactions Kinetix AI world-model system with base, action, and evaluation modules for predicting future states, generating candidate actions, and evaluating task progress and contact safety before execution; independent coverage also reports LLM integration Klara OS with on-device and cloud learning across sessions; core head, body, and tail behaviors work with no internet required Klara OS with on-device-by-default AI processing, user-enabled cloud features, session learning, and a Soul Chip identity/personality memory layer Knowledge-hypergraph autonomous decision-making architecture with local on-device inference; UESTC reports over 97% physiological-measurement accuracy and over 92% emotion-recognition accuracy in validation Kynooe describes no-code AI-powered interaction, app/web workflows, on-device AI face tracking, and a planned open-source SDK/Robot Hub ecosystem; underlying models and compute hardware are not officially disclosed. Large language model integration, visual perception systems, autonomous locomotion Laser and camera-based navigation with sensor fusion for room mapping and obstacle handling; Bosch does not disclose a named onboard AI model. Learning semantic engine with Ryzen 7 standard compute; optional AI module listed as Jetson Orin AGX (~275 TOPS) or RTX 5070. Exact production software stack is not officially disclosed. Learning-based control policy trained for side-by-side and in-line wheeled driving, with zero-shot deployment of behaviors including ground recovery and one-wheel balancing on hardware Lels Pro Dual-lidar Navigation With Aivi 3D Obstacle Avoidance And Horizon X5-based 10 Tops Object Recognition LeRobot-compatible runtime, MuJoCo simulation controller, MJLab reinforcement-learning training environments, ONNX/Torch policy execution, and simulation-parameter identification tools LG Physical AI combining Vision Language Model (VLM), Vision Language Action (VLA), and voice-based generative AI LLM-powered Reasoning With Wandercraft Motion-control Software Trained From Real-world Exoskeleton Movement Data LLM/RAG full-duplex interaction stack with edge deployment, facial recognition, lip-reading, ActionGPT motion generation, HIMUS 3D-SLAM, VectorFlux planning/control, and RTMOF motion control. Low-inertia dynamic manipulation research platform using torque control, trajectory acceleration feedforward, task-readiness impedance for catching, and real-time hardware learning for juggling patterns Low-power high-performance compute chip, robust locomotion-control algorithms, Chinese/English voice-command recognition, and owner/person recognition features; exact model stack not officially disclosed Lucid OS autonomy with Click-and-Clean job templates, saved/repeatable routes, NVIDIA edge compute, advanced vision, mapping intelligence, and zone-based safety controls. LySee 2.0 navigation combining RTK, VSLAM, and AI vision obstacle avoidance for wire-free mapping and route planning
M-P

73 entries

Single-robot components kept off the main scan path
MagicLab emotional interaction system and SAGE AI algorithm; AI voice interaction includes offline commands, intercom, music playback, and LLM conversations that may require a paid subscription Matic's official materials describe localized on-device intelligence for real-time 3D mapping, visual mess detection, automatic vacuum/mop mode switching, and obstacle-aware navigation. RTINGS and Vacuum Wars corroborate the five-camera, vision-first 3D mapping approach; exact model architecture and compute specifications are not publicly disclosed. MediaTek octa-core SoC + dual-core APU, LLM-powered Relationship Orchestration Engine Micropolis Microspot AI layer for real-time monitoring, fleet coordination, data analytics, and command-and-control integration across the company's robot ecosystem; exact M1.5 onboard compute configuration has not been publicly disclosed. Midea-developed humanoid robot technology stack; detailed compute, perception, and AI model specifications have not been publicly disclosed. Multimodal AI with long-term memory, contextual understanding, multi-model person/pet detection Multimodal contextual AI with real-time eye tracking, gesture and facial-expression recognition, voice-tone understanding, AR environmental overlay, and on-device encrypted processing Multimodal conversational AI with emotion-aware interaction, customizable personality profiles, and long-term memory Multimodal interaction, contextual awareness, and long-term memory (exact AI stack not officially disclosed) MyMemo Engine with proprietary memory intelligence, empathetic dialogue, adaptive learning, and personalized behavioral learning NAOqi OS (Linux-based) Narmind Pro with TwinAI/VLM obstacle recognition; unlimited object recognition via on-device processing and cloud learning; adaptive risk-based obstacle avoidance Native VLA data acquisition and management workflow, fully open SDK/high-low-level access, and Python/C++ plus ROS1/ROS2 development support NEBULA AI system — reinforcement learning and imitation learning, semantic task processing, natural language commands NEBULA AI system (100 TOPS computing) — visual recognition, visual SLAM, multimodal interaction, hand-eye coordination NEURA Adaptive AI / in-house developed AI with language model, computer vision, reinforcement learning, sim-to-real training, and Neuraverse skill integration NeuroNav AI for real-time navigation and obstacle avoidance; onboard AI identifies stain type and selects cleaning method; UV + RGB camera stain verification after cleaning No-code virtual spiking-neuron programming environment: users connect neurons, synapses, sensors, and motors in an app to create real-time reactive behaviors without LLMs or traditional code. NoshOS proprietary culinary AI, trained on thousands of cooking techniques and cuisines; natural-language recipe generation Not officially disclosed for this model. NVIDIA GPU, Multi-LLM integration (agnostic), Vision-Language Models (VLM), VSLAM navigation NVIDIA Isaac GR00T XX foundation model, Aura AI contextual intelligence, Neuraverse fleet-learning OS with shared skill propagation NVIDIA Isaac Sim for training; autonomous navigation NVIDIA Jetson AGX Orin 32GB onboard compute rated at 200 TOPS, with Galaxea documentation for ROS2-based development workflows and teleoperation NVIDIA Jetson AGX Orin 64GB + 1TB SSD NVIDIA Jetson AGX Thor T5000 with Blackwell GPU, 2,070 FP4 TFLOPS, 14-core Arm CPU, and 128GB unified memory NVIDIA Jetson NX, dual co-processors, 8GB RAM, 16GB storage NVIDIA Jetson Orin (200 TOPS) NVIDIA Jetson Orin (5x previous-gen compute); Level 4 autonomy with latest AI architecture for ultra-fast navigation decisions and collision avoidance NVIDIA Jetson Orin 64G + 16-core CPU NVIDIA Jetson Orin NX (157 TOPS) Nvidia Jetson Thor As The Core Domain Controller NVIDIA Jetson Thor edge compute with NVIDIA Isaac Sim and Isaac Lab simulation/training workflows, real-time reasoning, and simulated-plus-real-world reinforcement learning for industrial tasks. NVIDIA Jetson Thor, VLA Multimodal Model, NVIDIA Isaac GR00T NVIDIA Orin + RK3588 dual-processor autonomy stack for real-time SLAM, 3D reconstruction, object recognition, dynamic obstacle avoidance, path planning, and predictive following; up to 275 TOPS. NVIDIA Orin AGX compute rated at 275 TOPS, a dedicated real-time control processor, and Lumos' multimodal perception and decision-making stack; model details are not publicly disclosed. Nvidia RTX GPU modules (3x compute vs Figure 01), OpenAI speech model NVIDIA Xavier 32GB + 2TB NVMe SSD, perception-aided autonomy Official copy describes ongoing intelligent learning and task optimization for coordination with KEENON's service-robot ecosystem. Humanoid.guide's non-manufacturer-verified profile adds multimodal fusion perception, spatial awareness, and domain-specific knowledge graphs. Official English product page lists Doubao and iFlytek language-model support, while the official Chinese page also lists Qianwen/Qwen; Hobbs W1 combines these AI interaction options with bionic facial expression control and autonomous laser-SLAM navigation Official launch copy cites customizable AI computing; independent coverage says Astribot trains the system largely from human demonstration data. Detailed model and autonomy specifications are not officially disclosed. Official materials confirm autonomous and teleoperation modes for real-world task execution; independent sources describe task-planning AI and autonomous navigation, but Zeroth has not published compute or model specifications. Official materials describe advanced intelligence, emotionally responsive interaction, open behavioral customization, and embodied-intelligence creation use cases; exact AI stack not officially disclosed Official materials describe voice interaction, personality-style responses, smart motion, and app-based EDU programming support; independent launch coverage describes AI-powered conversation, but the exact model/provider is not officially disclosed. Official product page describes multimodal adaptive following plus multimodal fusion of visual, gaze, and gesture inputs for attention and emotion understanding. OmniSense 3.0 with binocular AI vision and 3D LiDAR; 300+ obstacle recognition; autonomous route planning On-device face, person, and object detection through the Grove Vision AI camera module, optional custom local models, and optional online analysis through the open-source XiaoZhi AI platform; programmable in Python with online and offline operating modes. On-device LLM with local visual processing for speech, gesture comprehension, face recognition, emotion cues, diary memories, and real-time companion behavior On-device OmniSense vision-language-action (VLA) model Onboard AI with open ROS-module development for navigation, manipulation, interaction, and customer-specific automation pilots Onboard IC chip running Yukai Engineering's randomized behavior algorithm for lifelike glance and reaction patterns Onboard Navigation With Systematic Mowing Patterns Onboard subject tracking and automatic highlight-editing workflow; detailed perception model, compute, and safety stack have not been publicly disclosed. Open-source autonomy stack (ROS 2 + Python SDK) Open-source Python SDK with Hugging Face model/app integrations for speech, vision, and conversational behaviors Open-source Python SDK with ROS1/2, LeRobot, Pinocchio, and depth-camera visual grasping support; Isaac Sim simulation support is listed as in progress. Open-source ROS 2 and Python SDK with reference autonomy demos for mapping, navigation, 3D SLAM, data collection, and VLM grasping; IEEE Spectrum reports Intel NUC 15 plus NVIDIA Jetson Orin NX onboard compute. OpenRTP platform (OpenRTM-aist, OpenHRP3), Linux-based control system ORBIT describes perception for structured task spaces and a path toward autonomy through teleoperation, imitation-learning experiments, and autonomous task policies; exact autonomy stack not officially disclosed Physical AI stack using AI-powered vision, real-time decision-making, robotic manipulation, LocusONE fleet orchestration, and NeuraGrasp AI-driven grasp planning for variable SKU handling. Piaggio Fast Forward leader-following and navigation software for operator identification, obstacle avoidance, speed matching, and pedestrian-aware following; detailed AI stack not publicly disclosed. PL-Universe SPDAA parallel acquisition framework, InduThread-VLA industrial VLA architecture, reusable autonomous decision-making, and clustered 'Super Brain' planning with local 'Smart Cerebellum' control PoolNavi AI-driven path planning with 360° AquaScan underwater LDS 3D mapping, adaptive debris detection, dynamic route optimization, and intelligent suction adjustment PoolSense 2.0 maps pool environments and generates adaptive multi-pattern path planning; Dreame has not disclosed model architecture or onboard compute details. Programmable action and sound-effect workflows through Robosen Studio and Robosen Hub; no general-purpose onboard AI assistant has been officially disclosed. Proprietary AI perception for autonomous clothing pickup, washing, drying, and retrieval; model architecture not publicly detailed. Proprietary Emotional AI With More Than 4 Million Emotional Possibilities And Local Voice-feature Processing For Owner Recognition Proprietary neural network architecture by Matrix Super Intelligence with zero-shot generalization; visual-tactile feedback loop for material, shape, and grip-stability assessment Proprietary VLA models (GraspVLA, GroceryVLA, TrackVLA) with NVIDIA Isaac Sim training pipeline Proprietary/embedded Realbotix AI; official materials describe third-party integrations for local AI applications and cloud providers including ChatGPT/OpenAI and DeepSeek, with Llama, Gemini, and Claude rollouts planned PsiBot describes ψ-SynRobot as the data-driven core carrier for its embodied-intelligence stack, combining task execution, field data capture, model iteration, and capability upgrades. Humanoid.Guide reports external LLM/model deployment support through a Thor compute platform, but PsiBot has not published a public software spec sheet for the robot. PUDU proprietary navigation algorithms with dynamic map updates, VSLAM + LiDAR SLAM localization, AI-driven multi-level obstacle avoidance, and multi-robot self-networking Pudu SLAM (dual LiDAR + Visual SLAM navigation)
Q-T

65 entries

Single-robot components kept off the main scan path
Qualcomm Dragonwing AI processor (deep learning) Qualcomm QCS605 (x2) + Qualcomm SDA660 + Amazon AZ1 Neural Edge Qualcomm Snapdragon 820 and 410 processors; Windows IoT Core main OS and Android 8 for navigation/computer vision; computer vision includes facial recognition and deep-learning AI using Qualcomm Snapdragon Neural Processing Raspberry Pi 4B (4 GB RAM) running Ubuntu 18.04 + ROS Melodic, OpenCV vision, inverse-kinematics and inverted-pendulum gait control, with API support for DeepSeek and Doubao multimodal LLM interactions Raspberry Pi 4B/5 host running Debian Bookworm with open-source Flask/WebRTC and JupyterLab tutorials; OpenCV and MediaPipe demos cover color recognition, automatic targeting, face detection, object recognition, gesture control, and vision line tracking, while the ESP32 sub-controller handles real-time motion/sensor loops. RDK X5 compute module; AMP anthropomorphic gait algorithm; ROS 2 deployment stack; IsaacLab reinforcement-learning training workflow with Sim2Sim/MuJoCo transfer Real-time perception and intelligent decision-making for path planning, obstacle avoidance, security patrol, factory operation, and harsh-environment industrial tasks Reinforcement learning combined with model-predictive control for basketball shooting, dribbling, and court movement Research autonomy stack for outdoor urban operations, building exploration, search behaviors, behavior cloning, simulation, perception, and VR teleoperation; exact compute and model details have not been officially disclosed. Research Platform For Service-robot Autonomy And Assisted Teleoperation In Home Environments RK3588 dual compute + NVIDIA Orin NX 157 TOPS (Ultra) RK3588 onboard compute platform with about 6 TOPS NPU compute for real-time inference, state estimation, motion scheduling, and safety monitoring; larger training and foundation-model workloads can be extended through external compute infrastructure. RobCo Physical AI with self-learning policies for Level 3/4 autonomous industrial workflows RoboForce Physical AI / robot foundation models for industrial deployment, trained with proprietary real-world industrial data plus simulation/world-model data; official March 2026 update says the stack uses NVIDIA Jetson Thor at the edge, Isaac Sim/Lab, Cosmos, and OSMO. Robopoet describes a MEM multimodal emotion model, EchoChain long-term memory system, and GrowMe personality-growth system that adapts Fuzozo's personality and responses through interaction; exact model providers, onboard compute, and cloud/on-device split are not officially disclosed. Roborock AI Algorithms For Wheel-leg Mobility And Environmental Understanding Robot Cloud API/CLI for high-level agent control; Asimov API for low-level robot data and commands; Virtual Asimov and real-time teleoperation apps; on-robot pre-trained RL walking policy; custom AI agents embodied via Cloud API Robot Operator Model-1 (ROM-1), Transformer-based control, imitation + reinforcement learning Robot PC with an Intel module (optional RK3588); AI compute varies by edition from NVIDIA Orin NX 16G to AGX Orin 64G, with custom upgrades noted by EngineAI ROBOX Navigation System with 2D mapping, 3D localization, user detection/tracking, obstacle avoidance, path planning. ASR, NLP, TTS engines for voice interaction. Facial recognition. ROS 2 + Python SDK, compatible with Hugging Face LeRobot, Pollen-Vision for perception ROS 2 API with ros2_control, full reinforcement-learning pipeline, mjlab open-source physics simulation, and optional NVIDIA Jetson AI Kit ROS 2 development stack with MoveIt 2, Nav2, ros2_control, PAL Web GUI, Docker PAL SDK image, RViz plugins, MuJoCo and Gazebo simulation support, and platform hooks for perception, teleoperation, embodied AI, and data collection workflows ROS-based (Ubuntu LTS, Real-Time OS) ROS-based autonomy with MoveIt!, SLAM navigation, whole body control, facial and speech recognition ROS-based stack with Python/C++/Java APIs; RD-V2 variants include Intel NUC i5/i7 or NVIDIA Jetson AGX Orin options ROS-based; real-time ros_control loop at 200 Hz; MoveIt! for motion planning; Whole-Body Control RTAB-Map SLAM with the D455 RGB-D camera, parallel wire-driven and wheeled-legged controllers, and an SMACH state machine for mode transitions; reported demos still include partial operator input rather than fully autonomous task execution. Screen-aware embodied AI assistant with audio-visual intent detection, attention/state awareness, emotion-aware interaction, clipboard/screen context, and productivity workflow integrations Self-developed embodied-intelligence model for task understanding and autonomous action-path planning; GigaBrain 1 model planned for Q3 2026 Self-developed motion control system; supports drag-and-drop graphical programming and voice interaction Self-developed multimodal perception and intelligent path-planning stack; the company also markets personality development based on ongoing household interactions. Self-evolving embodied AI with reinforcement learning from human feedback, adaptive dirt-sensitive cleaning, obstacle avoidance, and targeted mess hunting Semi-autonomous with human operator interface; FPGA-based 200Hz control loop Sensor-fusion navigation combines 3D LiDAR, camera vision, and VIO for centimeter-level positioning, auto mapping, route planning, AI obstacle avoidance, and multi-zone lawn settings. Sentisphere environmental perception with 360° 3D LiDAR, VSLAM, and Vision-LiDAR Fusion obstacle avoidance Sharp CE-LLM conversational AI running on a Qualcomm Snapdragon 662 octa-core processor with 3GB RAM and 32GB storage. Sim2Real learning for gait and hand movement; NeRF-based real-time 3D mapping/localization; dynamic navigation; Large Language Models for cognitive mapping and advanced task execution Six-engine emotional AI system; personality develops over time based on interactions SmartMap navigation that learns pool shape and tailors cleaning paths, plus JetIQ directional-jet maneuvering for stairs and curved pool sections. Smartnavi Intelligent Navigation System With IMU-based Route Optimization For Efficient Coverage SmartVision AI for grass detection, boundary recognition, and obstacle avoidance SolarSeeker sunlight-seeking recharge behavior plus dual-sensor obstacle avoidance, smart edge cleaning, anti-stranding alerts, app scheduling, and remote steering for autonomous surface skimming SonicSense ultrasonic obstacle avoidance and SolarTrack light tracking for autonomous full-surface skimming, app scheduling, Smart Auto Parking, and one-tap recall SonicSense ultrasonic obstacle detection and avoidance, optimized S-shaped cleaning path planning, automatic zone adaptation for floor/walls/waterline/platforms SonicSense ultrasonic obstacle detection and avoidance, optimized S-shaped cleaning path planning, automatic zone adaptation for surface/floor/walls/waterline/platforms Sony proprietary deep learning AI (cloud + edge) Sony proprietary; face/voice recognition, emotional behavior system Specific onboard compute has not been officially disclosed; AGIBOT positions G2 Air within its embodied-AI deployment stack for task execution and real-time data collection. Split edge/cloud AI architecture: on-device perception and real-time control with cloud-based multimodal reasoning, memory, emotion modeling, and dialogue planning Starship Level 4 autonomy (machine learning, feature detection, robotic mapping) StarSight Autonomous System 2.0 with AI object recognition StarSight Autonomous System 2.0; 300+ object type recognition; VertiBeam lateral avoidance Sunday ACT-1 robot foundation model trained with the company's Skill Capture Glove / Skill Transform data pipeline Symbolic AI, neural networks, expert systems, NLP, adaptive motor control, cognitive architecture (SOUL), CereProc TTS Syncere ClearTouch home-robotics stack plus Personalization Band preference learning; the company has not publicly detailed the underlying model architecture. Terrain-aware posture and gait adjustment with autonomous navigation; omnidirectional obstacle avoidance and point-cloud surround view are listed as future OTA-enabled features. TerraVision 2.0 combines front/rear camera coverage, RTK-assisted positioning, 3D visual sensing, AI semantic obstacle perception, and one-click auto mapping for wire-free lawn navigation. Tesla Autopilot-derived Neural Network Tesla-developed Neural Network Ti0 vision-language-action model trained on Data Factory 1 fleet data; Tutor uses remote VR proprioceptive teleoperation, human feedback, and iterative policy retraining to improve Sonny's manipulation skills. Tombot describes autonomous robotic-pet behavior, touch-aware reactions, voice-command responses, and upgradeable behavior software; exact onboard compute, model provider, and cloud dependency have not been officially disclosed. Tri-Fusion positioning (360° LiDAR + NetRTK + dual-camera AI vision) with a 10 TOPS AI processor Tritium AI with default integrations for OpenAI ChatGPT, OpenAI Whisper, and Amazon Poly; custom integrations available; Tritium Roles supports purpose-driven behaviors. Truecourse Dtof LiDAR Mapping Plus 3D Matrixeye Perception And Real-time Obstacle Avoidance With Active Binocular Infrared And RGB Cameras
U-Z

27 entries

Single-robot components kept off the main scan path
Ubtech AI Platform UBTECH BrainNet 2.0 with Co-Agent industrial agent system for task planning, tool use, and anomaly handling UBTECH embodied AI / service-robot interaction stack; detailed C1 software specifications not officially disclosed UBTECH interaction stack with voice, face/object recognition, and balance control UltraSense AI Vision with 5 TOPS chip; recognizes 200+ obstacle types; autonomous path optimization UltraView 3.0 navigation with 360° 3D LiDAR, AI dual vision, AI-assisted auto mapping, obstacle avoidance for 300+ obstacle types, and U-shaped path planning UniFlex (imitation learning), UniTouch (tactile perception model), UniCortex (long-sequence task planning), multimodal semantic keypoints Unitree Reinforcement Learning Engine Unveiled alongside MagicLab's Magic-Mix world model; public sources do not yet disclose an X1-specific compute stack or onboard model specification Up to 2070 TOPS (Jetson AGX Thor optional); Intel Core i5/i7 onboard VinMotion has not published a full AI or compute specification. CNET reports that Qualcomm used Motion 2 to demonstrate its Dragonwing IQ10 robotics architecture at CES 2026, while independent robotics profiles describe on-device motion planning, real-time balance control, and human-in-the-loop supervision. Vision Fsd Camera-based Mapping And Obstacle Detection VisionPath AI navigation with dToF mapping, Cognitive AI debris detection, AI Navium autonomous scheduling, intelligent obstacle avoidance trained on 2+ years of real-world pool data across US, Europe, and Australia Viveka Decision Core AI, described by iHub as a VLA orchestration engine for contextual world understanding, multimodal reasoning, long-horizon task planning, natural-language command execution, and self-improving execution; detailed onboard compute is listed only as upgradeable AI compute. Volta Lawn Intelligence combines computer vision, GNSS, IMU data, a hex-cell lawn model, per-lawn learning, and fleet intelligence for adaptive mowing and lawn-health analysis Watchbot Intelligence Perception And Navigation Stack With Six-times More Onboard Computing Capacity Than The Prior Generation Weave AI (weekly model updates, learning from corrections) WIN-SLAM 5.0 path planning with dynamic obstacle avoidance and edge-aware cleaning logic Wire-free Epos Navigation With Optional AI Vision Object Identification And Night-time IR Support Wise KaiWu embodied AI platform with world model, embodied VLM, cross-ontology VLA models, perception-decision-execution loop, autonomous navigation, task planning, and multi-agent coordination WorkGPT multimodal interaction stack, LinkCraft motion creation, AimRT framework Xiaomi AI Platform Xinghai large model + DeepSeek integration for scene understanding and appliance coordination XPENG Turing AI Chip (3,000 TOPS), 30B parameter AI model, reinforcement learning locomotion YARP middleware + open-source ML frameworks Zerith self-developed embodied-intelligence algorithms for perception, manipulation, and task execution; model architecture details are not officially disclosed. ZonePilot AI Vision for real-time pool mapping, debris identification, step detection, and adaptive path planning
0-9

16 entries

Single-robot components kept off the main scan path
10 TOPS AI platform with LiDAR and vision fusion for automatic mapping, path planning, 300+ obstacle recognition, and smart cliff protection 128 TOPS onboard AI compute, self-developed spatial foundation model, on-device spatial intelligence, physical-space agent behavior, generative actions/dances, and large-language-model voice system 1x Embodied Intelligence 2× Intel i7 (8th gen, 6-core) edge computing 360° LiDAR and dual-camera AI vision for mapping, path planning, and 300+ obstacle detection 360° LiDAR plus dual-vision navigation with centimeter-level positioning, autonomous mapping, and AI obstacle detection for 1,000+ obstacle types 6 TOPS NPU (int4/int8/int16/FP16/BF16/TF32), Cortex-A76×4 + Cortex-A55×4 CPU, Mali-G610 GPU; NVIDIA Isaac Sim for RL training, imitation learning via leader-follower system 8-core body processor plus 8-core head processor with 10 TOPS head-module compute; optional NVIDIA Jetson Orin 40–100 TOPS module 8-core CPU + GPU; optional NVIDIA Jetson Orin (40–100 TOPS, EDU only); UnifoLM multimodal LLM 8-core CPU with 100 TOPS AI processor; MagicLab navigation algorithms, motion-control self-learning, 6D visual servoing, full-body imitation learning, and Atomic Myriad scene-model stack 8-core CPU, AI simulation-trained gaits, optional NVIDIA Jetson Orin (40–100 TOPS) 8-core High-performance Cpu 8-core high-performance CPU (optional NVIDIA Jetson Orin for EDU) 8-core high-performance CPU + optional NVIDIA Jetson Orin NX (EDU) 8-core high-performance CPU; MagicLab describes IL/RL-based human-like walking, rapid full-body motion learning, 360° perception, autonomous navigation, and multimodal dialogue, with an optional high-computing module on the development version 8-core Horizon Sunrise Series Cpu With Onboard Autonomous Navigation And Perception Stack

Understanding AI Components

AI is the layer that turns perception and rules into behavior. On ui44, this route is most useful when two robots look similar on paper but seem to behave very differently in navigation, interaction, or autonomy. AI labels can be noisy, but they still help explain planning quality, local versus cloud tradeoffs, and how much the robot can adapt beyond a fixed script.

The ui44 database tracks 352 ai components used across 354 robots.

How it works

AI in robotics is a chain, not a magic box. Sensor interpretation, world modeling, planning, and control all have to line up. That is why flashy AI branding without corresponding hardware, update cadence, or behavioral evidence often deserves skepticism.

Evolution

Robots moved from hard-coded control logic into SLAM-driven autonomy, then into richer perception models, object recognition, and increasingly multimodal reasoning stacks. The most recent shift is toward local AI acceleration paired with cloud augmentation rather than a purely one-sided architecture.

Evaluation Guide

What to check and what to watch for when comparing options

What to evaluate

Look for evidence of what the AI actually changes: mapping, obstacle handling, language interaction, personalization, planning, or manipulation. Then check whether those changes are on-device, cloud-based, or hybrid. The operating model matters as much as the label.

Deployment realities

AI quality is constrained by compute, thermals, network reliability, and data policy. A robot that depends heavily on cloud reasoning may feel less stable in weak-connectivity homes, while a fully local stack may trade some flexibility for consistency and privacy.

What's changing

The clearest trend is stronger on-device acceleration paired with better multimodal models. Buyers increasingly care about whether the robot can remain useful offline, update meaningfully over time, and explain behavior through real product outcomes instead of vague AI branding.

Frequently Asked Questions

AI technology
What does AI actually tell me on a robot page?

Ideally it tells you what class of behavior the robot can deliver, not just what buzzword the brand prefers. Good AI evidence should connect to navigation, interaction, adaptation, or planning outcomes you can recognize elsewhere on the profile.

Why does local versus cloud AI matter so much?

Because it changes latency, privacy, reliability during outages, and long-term dependency on the vendor. The same feature can feel very different depending on where the model actually runs.

How should I sanity-check an AI claim?

See whether the product story stays coherent across sensors, compute hints, update cadence, and real-world use case framing. When the rest of the profile does not support the AI story, treat the label cautiously.

Using this directory
What does robot count tell me on this route?

It is a browse signal, not a quality score. Higher counts usually mean shared comparison anchors. Lower counts often mean proprietary or more signature-heavy technologies that need product context before they become meaningful.

How should I compare similar labels?

Use the component layer for evidence, the robot page for context, and Compare for decisions. Shared labels do not automatically mean identical behavior.

When should I leave this route?

Leave once the question becomes product fit instead of technology meaning. The component layer should narrow your attention, then hand you off to the product routes where price, form factor, deployment fit, and broader system design matter.

Continue exploring

Glossary Sensor Actuator Connectivity Voice Assistant All robots