Can Home Robots Open Doors?

That sounds almost silly because people do it dozens of times a day. For a robot, though, a door is a compact version of the whole home-robot problem. The machine has to see the handle, understand which way the door moves, place its base where the arm can work, apply force without overdoing it, move its body as the door swings, and confirm that the path is safe before walking through.

That is why door opening is a better buyer benchmark than many flashy demos. It is not just a gripper trick. It is a test of whole-body manipulation: arms, legs or wheels, sensors, balance, contact control, and decision-making all working together in a human space.

Why door opening is such a hard home-robot test

A useful home robot does not merely need to touch a handle. It needs to solve a chain of small problems that are easy to hide in a product video:

• Find the actionable part: lever, knob, pull handle, sliding door groove, latch, pet gate, cabinet pull, or screen-door edge.
• Know the door geometry: hinge side, swing direction, whether the robot is on the push side or pull side, and whether furniture blocks the arc.
• Position its body: close enough for the arm, far enough not to be hit by the door, and stable enough to apply force.
• Handle contact: press, rotate, pull, or push while sensing force, slip, friction, and a latch that did not release.
• Move through the action: continue walking or rolling while the arm remains connected to the handle or door surface.
• Stop safely: avoid fingers, pets, people, glass, walls, and a door that is locked, swollen, spring-loaded, or intentionally child-proofed.

This is why a robot can have an impressive arm and still be bad at doors. A stationary arm can open a lab door if the handle and path are prepared for it. A home assistant has to deal with doors that were never designed for robots.

What is whole-body manipulation?

Whole-body manipulation means the robot uses more than its wrist and fingers to complete a physical task. Humans do this constantly. We lean into a heavy door, brace one hand against a wall, twist our torso while pulling a handle, step backward as the door opens, and shift weight without thinking about it.

1X uses door opening as one of the examples for its Redwood AI work. The company says Redwood is a vision-language transformer for NEO that performs end-to-end mobile manipulation tasks such as retrieving objects, opening doors, and navigating around the home. More importantly, 1X says Redwood jointly controls locomotion and manipulation, including pelvis pose, walking commands, arms, and hands. Its own explanation calls out bracing and leaning behaviors, including bracing a hand against a wall when pulling open heavy doors.

That is the right framing. A humanoid arm is not enough if the body cannot move with the door. The same is true for wheeled mobile manipulators: the arm, base, and planner have to act as one system.

In ui44's database, 1X NEO is tracked as a home-focused humanoid in pre-order at $20,000, about 167 cm tall and 30 kg, with roughly 4 hours of battery life, RGB/depth sensing, tactile skin, a microphone array, and gentle-manipulation claims. 1X's order page goes further, listing an arm payload of 18 lb, carry of 55 lb, lift of 154 lb, a soft body, IP44 body splash protection, and scheduled remote Expert Mode for complex tasks.

Those numbers are meaningful, but the door benchmark asks a stricter question: can the robot combine them reliably in an ordinary hallway after the furniture moves and the handle is not where the training clip expected?

Why arms alone are not enough

Door opening is contact-rich. The robot cannot solve it purely by recognizing an object in an image. It has to discover how the object reacts while force is being applied.

A 2023 paper in the Journal of Robotics and Mechatronics is useful here because it separates the door problem into push-opening, pull-opening, push-closing, and pull-closing operations with a mobile manipulator. The authors had to reason about force relaxation, admittance control, and how the mobile base should move while the arm stays in contact. That is a long way from "the robot has a hand".

The broader research picture says the same thing. The HomeRobot open-vocabulary mobile manipulation benchmark defines the home-robot challenge as bringing together perception, language understanding, navigation, and manipulation. Its real-world baselines on Hello Robot Stretch achieved 20% success on pick-and-place style tasks. Doors add harder contact, higher forces, and a moving obstacle attached to the building.

MERL president Anthony Vetro made the reliability point bluntly in an April 2026 interview with Robotics & Automation News: robots are making progress in manipulation and force control, but they are still not at human-level reliability for contact-rich tasks. He tied progress to tighter integration of perception, physics-based reasoning, and real-time response under uncertainty.

That is the core buyer lesson. A robot that opens a door once in a polished demo is interesting. A robot that can decide not to force a jammed door, ask for help, and avoid damaging your home is closer to useful.

Which current robots look closest?

The strongest candidates are not always the tallest humanoids. A door-opening robot needs reach, payload, force sensing or compliance, body mobility, and a software stack that can recover when the action goes wrong.

Stretch 3 deserves special attention because it is honest about what it is. The official product page calls it a fully integrated mobile manipulator, available for $24,950, with a 33 × 34 × 141 cm body, 24.5 kg weight, 2–5 hour runtime, 2 kg payload, ROS 2, Python SDK, autonomy demos, and several teleoperation modes. That does not mean it will casually open every door in your house. It means researchers and developers can actually work on the integrated problem in real rooms.

Toyota HSR is the older but still revealing example. Toyota describes it as a home-living and care robot for helping elderly and disabled users live more independently. The official specs list a 430 mm body diameter, 100.5–135 cm body height, about 37 kg weight, 600 mm arm length, and objects up to 1.2 kg. It can pick objects up from the floor and retrieve items from shelves, and it supports remote operation by family or caregivers. That is not a door-opening consumer product, but it shows how long serious home manipulation has been treated as an assistive, shared-control problem.

How to judge a door-opening demo

The next time a company shows a humanoid opening a door, watch for details that are usually more important than the final moment.

1. Is it a push door, pull door, knob, lever, or sliding door?

A push door with a lever is easier than a pull door with a knob and spring closer. A sliding patio door is different again. Good demos should state the door type instead of implying all doors are equivalent.

2. Does the robot keep contact while the base moves?

If the arm works but the base freezes, the robot may be solving only the first half of the task. Whole-body manipulation means the body moves as part of the action, not after the action is already done.

3. Does it sense force, or is it replaying a path?

A replayed path can look convincing until the latch sticks or the door is already partly open. Useful home robots need compliance, force awareness, or a safe fallback when contact does not match the plan.

4. What happens when the attempt fails?

This is the most important question. The right behavior may be to stop, explain what happened, ask a person, or switch to approved teleoperation. The wrong behavior is to yank harder until something breaks.

5. Can it respect household rules?

Opening a door is not always allowed. Bedrooms, bathrooms, exterior doors, medicine cabinets, garages, and pet gates all have privacy and safety meaning. A home robot needs room-level permissions, guest modes, logs, and clear refusal behavior.

Should buyers care about door opening now?

Yes, but not because everyone needs a robot to open doors on day one. Buyers should care because door opening exposes whether a robot is being built as a real physical assistant or as a moving screen with arms.

For most homes in 2026, the practical answer is still cautious:

• If you want a developer platform for manipulation research, Stretch 3 is the most concrete home-relevant option in ui44's database.
• If you want an early home humanoid, NEO is the most explicit consumer-facing door-opening claim, but it is still a pre-order product and should be judged by real owner evidence, not launch-page language alone.
• If you want an affordable humanoid to experiment with, Unitree G1 is available and comparatively inexpensive, but it is still a serious research machine with safety limits.
• If you want a polished appliance, door-opening humanoids are not there yet.

The near-term milestone is not "a robot opened one door once." It is a robot that can open simple allowed doors repeatedly, explain when it cannot, avoid unsafe rooms, recover from contact errors, and preserve the home around it.

That is why door opening matters. It is a small chore that reveals the big truth: home robots become useful when navigation, manipulation, force control, permissions, and humility all show up in the same machine.