Home Robot Handovers: Safe Object Passing Explained

That distinction matters because home-robot demos are moving from waving and walking toward manipulation: towels, cups, bottles, drawers, packages, remote controls, laundry, and medicine bottles. The moment a robot gives something to a person, the task stops being only about arm strength. It becomes a shared action between a moving machine and a moving human.

Addverb's recent work with IIT Gandhinagar is a useful reminder of how much is hidden inside the word "handover." The research was not a consumer-home trial. It used robot arms and Addverb hardware to study robot-to-robot transfers of boxes and bottles. But the failure mode is exactly the one home buyers should care about: if the giver and receiver do not arrive together, face the object in the right orientation, and adapt when a path is blocked, the transfer can become a drop, a collision, or an unsafe reach.

For buyers, the question is not "does this robot have hands?" It is what proof shows the robot can pass, receive, refuse, and recover around real people?

What is a robot handover?

A robot handover is a physical transfer of an object between two agents. One is the giver, one is the receiver. Either side can be a person or a robot.

That sounds simple until you break down the pieces. The robot needs to detect the object, choose a grasp that leaves room for the other hand, approach at a speed that does not startle the person, orient the object correctly, sense whether the receiver is ready, release at the right moment, and stop safely if anything changes.

A home handover can be as ordinary as a robot bringing a water bottle from the kitchen to a sofa. It can also be safety-critical: a medication bottle, a phone, a cane, a hot mug, or a fragile item. In those cases, a single edited demo is not enough evidence.

This is why robot-handover research often studies timing, grip force, visual tracking, haptic cues, and motion planning together. A 2021 Frontiers in Robotics and AI paper described handover control as depending on visual and haptic perception, robot motion control, and grip-force control so the exchange feels intuitive, fluid, and safe. In plain English: the robot has to know when to hold on, when to let go, and how to adapt if the human does something slightly different from the plan.

What did Addverb and IIT Gandhinagar show?

Addverb's post, "Two Arms, One Handover," describes work by Debojit Das at the IIT Gandhinagar Robotics Laboratory in collaboration with Addverb Technologies. The team built a control framework for synchronized robot handovers. The core idea is a shared sense of progress: instead of two arms independently moving to a fixed point and hoping they line up, the arms adjust their pace together.

The framework uses an asynchronous phase while each arm moves independently, then a synchronous phase where the arms are phase-locked in position and orientation. That second word is important. Position alone is not enough. If two grippers arrive at the same point but face the object from the wrong angle, the handover can still fail.

Addverb says the system was tested on two hardware setups: two Syncro 5 cobot arms with parallel-jaw grippers, and Addverb's humanoid robot with a dexterous hand on one arm and a parallel-jaw gripper on the other. It tested box transfers, bottle transfers, and obstacle cases where a path changed mid-motion.

The most useful numbers are the ablation results. With synchronization coupling active, Addverb reports 5 out of 5 successful transfers for both box and bottle trials. Without coupling, box transfers failed entirely, and bottle transfers succeeded only 1 out of 5 times. That is the kind of difference buyers should look for in home-robot evidence: not just a successful clip, but a comparison showing what makes the success repeatable.

It is also important not to overread the result. This was not a public home deployment, not a medicine-handoff study, and not proof that a humanoid can hand objects to a child or older adult unsupervised. It is evidence that timing and orientation are not polish details. They are core safety mechanics.

Which ui44 robots are closest to useful home handovers?

The ui44 database helps separate three very different categories: research and assistive mobile manipulators, home-focused humanoids, and consumer products with small object pickup.

The table shows why "has arms" is too crude. Stretch 4 is not humanoid, but its home-sized footprint, ROS 2/Python stack, depth sensing, floor hazard sensing, and 2.5-4 kg arm rating make it a serious handover-research platform. Roborock Saros Z70 has a much smaller 300 g object limit, but it is actually shipping to consumers. Figure 03 has far more payload on paper and Figure now positions it for homes, but ui44 has no public price, purchase path, or repeated home handover evidence yet.

The best home-handover candidate is not necessarily the strongest robot. It is the robot whose manufacturer can document the object list, payload envelope, approach behavior, release cue, stop behavior, and recovery plan.

Why payload alone is not enough

Payload is easy to market. Handover safety is harder.

A robot might lift 10 kg, but that does not mean it can safely pass a 1 kg object to a person. Handover quality depends on the full chain: gripper shape, wrist orientation, arm compliance, object detection, release timing, human-intent sensing, and the robot's response when the person hesitates.

Consider three everyday examples:

• A water bottle needs orientation control so the receiver can grab the body or neck naturally. The robot should not present it cap-first into a palm.
• A folded towel is deformable. The robot may need to hold it without crushing, dropping, or hiding the grasp point.
• A medication bottle is light, but the consequence of dropping it or handing it to the wrong person is much higher than the weight suggests.

This is where Addverb's orientation result matters. A geometrically valid pose is not always a functionally valid handover. The object has to arrive in a pose that lets the receiver take it.

The same principle applies to home buyers using the ui44 robot comparison tool. A spec sheet can tell you payload, sensors, price, and status. It cannot by itself tell you whether the robot gracefully handles a person reaching early, pulling late, turning away, or saying "actually, put it on the table."

What should buyers ask before trusting a handover demo?

A useful handover demo should answer more than "did the robot pass the object once?" Here is the checklist ui44 would use before treating a home handover claim as meaningful.

1. What objects were tested?

A robot passing a rigid box is not the same as passing a soft towel, a slippery bottle, a mug with a handle, a phone, or a pill organizer. Ask for the object list and the failure list. The failure list is often more informative.

2. How many trials were run?

Five successful trials are not enough for a commercial safety claim, but they are better evidence than one cinematic clip. The useful number is success rate across varied people, objects, approach angles, lighting, and clutter.

3. What is the release cue?

Does the robot release when it sees the hand? When it senses pull force? When a voice command is heard? When a button is tapped? If the release cue is vague, the handover is not ready for unsupervised home use.

4. What happens when the person is not ready?

A safe home robot should be able to pause, hold, retract, set the object down, or ask for clarification. "It drops the object" is not recovery. "It waits forever with the arm extended" is not good recovery either.

5. Is the handover bounded by user permissions?

The robot should not hand every object to every person. Medicines, sharp tools, cleaning chemicals, alcohol, and personal items need identity, consent, or at least household rules. This is a product-design issue, not just a manipulation issue.

6. Can the user choose a fixed handoff point?

For early home robots, the safest pattern may be boring: bring the object to a tray, counter edge, table, wheelchair side shelf, or marked handoff zone. Direct hand-to-hand transfers are impressive, but a predictable placement point can be more reliable.

How close are home robots to safe object passing?

Closer than they were, but not close enough to trust the broadest claims without proof.

The encouraging part is that several pieces are now real. Addverb's work shows synchronized handover control on physical hardware. Stretch 4 gives researchers and assistive pilots a home-scale mobile manipulator with serious sensing and a clear payload envelope. 1X NEO is explicitly aimed at homes, and Figure 03 shows how much investment is going into humanoid manipulation. Roborock Saros Z70 proves that a consumer robot can ship with a small arm and object pickup, even if it is not a human-handover device.

The limiting part is that homes are hostile test environments. People sit, turn, hesitate, interrupt, hold pets, use walkers, leave toys on the floor, and ask for objects in awkward places. A robot that can coordinate with another robot arm in a lab still has to earn trust around unpredictable humans.

So the buyer-friendly answer is this: object handovers are a real capability to watch, but they should be treated as a bounded safety claim. Ask for the object list, the trial count, the release cue, the recovery behavior, and the deployment context. If a manufacturer cannot answer those questions, the robot may have an arm — but it has not yet proved it can safely hand you the thing you asked for.