Open Source Robotics

A curated collection of high-quality open source robotics projects, tools, and software to propel the robotics community forward.

Robot Handover

  • Also called: object handover, human-robot handover, handoff
  • Category: Control › Human in the loop
  • Tags: human-robot-interaction, grasping, collaborative-robots, physical-hri, manipulation

The physical transfer of an object between a robot and a person, either human-to-robot or robot-to-human, coordinated through motion, timing, grasp choice and force.

Definition

A robot handover is the joint action in which a robot and a person pass an object from one to the other. It is usually split into two directions: human-to-robot (H2R) handovers, where the robot takes an object the person is holding, and robot-to-human (R2H) handovers, where the robot offers an object for the person to take. Robot-to-robot exchanges also exist but are normally treated as a separate, fully controlled problem. A handover is a short piece of physical human-robot interaction: for a moment both partners hold the same object, and the transfer succeeds only if one lets go as the other takes the load.

A handover is commonly described in phases. In the approach or reach phase the giver presents the object and the receiver moves toward it; the robot must decide where to meet the person, at what height and with what timing. In the transfer phase both hands are on the object, and the giver releases its grip as the receiver takes the weight. Humans do this by sensing the change in load through the object, and robots often imitate it: in R2H handovers the robot opens its gripper when its wrist force-torque sensor or motor currents show that the person is pulling or supporting the object. In the retreat phase the giver withdraws and the receiver carries the object away.

H2R handovers add a perception and grasping problem. The robot has to track the object and the person's hand, often from a single depth camera with heavy occlusion, choose a grasp that does not collide with the person's fingers, and keep updating that choice as the hand moves. Reactive approaches segment the hand and the object in a point cloud, generate candidate 6-DoF grasp poses, discard those close to the hand and servo the gripper toward the best remaining grasp in closed loop. R2H handovers put more weight on legibility: offering the object at a natural height, with the handle toward the person and with predictable motion, so that the person knows when and where to take it.

Why it matters

Handing things over is one of the most common ways people and robots will share work in homes, hospitals and workshops. Handovers combine perception, grasping, motion planning and safety around a person in a single short action, and failures are immediately noticeable: dropped objects, pinched fingers or hesitant timing that people find uncomfortable.

Key applications

  • Collaborative assembly, with robots passing tools and parts to workers
  • Assistive robots handing objects to people with limited mobility
  • Surgical and laboratory assistance with instruments
  • Service robots in homes, retail and hospitality
  • Picking up objects people present to a robot during teaching

In practice

A collaborative arm with a parallel-jaw gripper and a wrist-mounted depth camera is set up to take tools from a technician. At each camera frame the system segments the person's hand and the tool from the point cloud, runs a grasp network on the tool points to get candidate 6-DoF grasp poses, and rejects any grasp whose gripper volume comes within a few centimeters of hand points. The arm moves toward the highest-scoring remaining grasp with a velocity-limited controller and switches to a slower final approach near the tool. It closes the gripper and then waits for the force-torque sensor to show that the tool's weight has moved to the robot before retracting. For the opposite direction, handing a tool back, the arm holds the tool at chest height with the handle toward the person and releases only when it measures a pull above a set threshold.

Common misconceptions

  • A handover is just a grasp of an object that happens to be held — the person's hand constrains which grasps are safe, the object keeps moving and the release must be coordinated with the partner
  • The robot should release as soon as its gripper closes — a release before the receiver has the load causes drops; well-designed systems release on a measured transfer of load
  • Faster handovers are always better — people prefer predictable, legible motion, and abrupt or very fast approaches lower comfort and perceived safety
  • Human-to-robot and robot-to-human handovers are the same problem reversed — H2R is dominated by perception and grasp selection, R2H by object presentation and release timing

Who should know this

  • Human-robot interaction researchers
  • Collaborative robot application engineers
  • Assistive and service robotics developers
  • Grasping and manipulation engineers working near people

Related technologies

  • Six-Axis Force-Torque Sensor
  • Hand Pose Estimation
  • 6-DOF GraspNet
  • MoveIt
  • ISO/TS 15066

Related terms

  • Physical Human-Robot Interaction
  • 6-DoF Grasp Pose
  • Grasp Planning
  • Perceived Safety
  • Cobot
  • Visual Servoing

Related projects

  • LeRobot: HuggingFace ML library for real-world robotics in PyTorch
  • Isaac Lab: NVIDIA robot learning framework: one task API across Newton, PhysX and Isaac Sim
  • NVIDIA Isaac GR00T: NVIDIA open foundation model for humanoid robot reasoning and skills
  • OpenVLA: 7B-parameter Vision-Language-Action model for robotic manipulation
  • Drake: MIT robotics toolbox for model-based dynamics, optimization, and control system design
  • OpenArm: 7DOF humanoid arm for physical AI research in contact-rich environments

This website requires JavaScript to function properly. Please enable JavaScript to explore our collection of open source robotics projects.