Open Source Robotics

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

Shared Autonomy

  • Also called: shared control, assistive teleoperation
  • Category: Control › Human in the loop
  • Tags: human-robot-interaction, teleoperation, assistive-robotics, intent-inference, control-arbitration

A control scheme in which a human operator and a robot act on the same task at once, with the robot inferring the operator's goal and blending its own assistance into the commands.

Definition

Shared autonomy describes systems in which a human and a robot control the same robot at the same time. The operator gives commands through a joystick, a brain-computer interface, a teleoperation device or another input, and the robot does not simply execute them: it estimates what the operator is trying to achieve and contributes its own actions toward that goal. The result sits between direct teleoperation, where the robot follows commands exactly, and full autonomy, where the human only sets the task.

Most formulations have two parts. Intent inference maintains a belief over candidate goals, such as which object the user wants to grasp, updated from the user's inputs and the robot's motion so far. Arbitration decides how the robot's assistance and the user's command are combined. Dragan and Srinivasa (2013) framed this as policy blending, in which the executed command is a weighted mix of the user's input and the robot's prediction, with the weight growing as confidence in the predicted goal rises. Other approaches treat the unknown goal as hidden state in a partially observable Markov decision process and choose actions that help across several plausible goals, or learn the assistance directly from data, for example by mapping low-dimensional user inputs to learned latent actions.

The central trade-off is between assistance and control authority. Aggressive assistance can complete tasks faster but acts on wrong guesses when goals are ambiguous, and users often report frustration when the robot overrides them. Many systems therefore keep the human able to override at any time, expose the robot's inferred goal through the interface, or adapt the level of assistance to each user over repeated interactions. Shared autonomy differs from traded control, where authority switches between human and robot in discrete episodes, and from physical human-robot interaction, where the human guides the robot through direct contact rather than through an input device.

Why it matters

Many useful robots cannot be fully autonomous yet, and many users cannot teleoperate a high-DOF arm precisely through a joystick or other limited input. Shared autonomy lets the robot handle fine positioning while the human keeps the decisions, which makes assistive arms, remote manipulation and difficult piloting tasks practical for non-experts.

Key applications

  • Assistive robot arms for people with limited upper-limb mobility
  • Remote manipulation with limited input devices or communication latency
  • Brain-computer interface and electromyography control of robotic devices
  • Driver assistance and collision avoidance in semi-autonomous vehicles
  • Drone and aerial vehicle piloting with automated stabilization and obstacle avoidance
  • Robot-assisted surgery with virtual fixtures and motion constraints

In practice

A wheelchair-mounted assistive arm is controlled with a two-axis joystick, so the user cannot command all six end-effector degrees of freedom at once. The software detects candidate objects on a table with a depth camera and keeps a probability for each one being the target, updated from how well the user's recent joystick motions point toward it. While the belief is spread across objects, the arm follows the user's commands almost directly. As one object becomes clearly the most likely, the controller blends in motion toward a pre-grasp pose for it, including the wrist rotation the joystick cannot express. The user can push against the assistance at any time, which lowers the probability of that goal and returns authority to them. Engineers tune the blending function and log override events to find where the inference is guessing wrong.

Common misconceptions

  • Shared autonomy means switching between manual and automatic modes — that is traded control; shared autonomy combines both sources of command continuously
  • More assistance is always better — assistance based on a wrong goal estimate slows users down, and many users prefer to keep more control even at some cost in speed
  • The robot must know the goal in advance — goal inference from the user's inputs is part of the problem, and good systems act sensibly while the goal is still uncertain
  • Shared autonomy is the same as teleoperation — in plain teleoperation the robot reproduces the operator's commands, while in shared autonomy it adds actions of its own

Who should know this

  • Assistive and rehabilitation robotics engineers
  • Teleoperation and human-robot interaction researchers
  • Interface designers for robots controlled by non-experts
  • Engineers building driver assistance or piloting aids

Related technologies

  • Policy Blending
  • Intent Inference
  • Virtual Fixtures
  • Brain-Computer Interfaces
  • Assistive Robot Arms

Related terms

  • Teleoperation
  • Physical Human-Robot Interaction
  • Partially Observable Markov Decision Process (POMDP)
  • Inverse Reinforcement Learning
  • Safety Filter

Related projects

  • SO-101 Open Robot Arm: 6-DOF robot arm for teleoperation and AI robotics research
  • XLeRobot: Practical dual-arm mobile home robot for $660 with <4hrs assembly
  • DimOS: Python robotics runtime with agents, SLAM and teleop, no ROS required
  • Mobile ALOHA: Bimanual mobile manipulation with 4 robot arms on a wheeled base for teleoperation
  • GEAR-SONIC: NVIDIA's whole-body controller for the Unitree G1, trained to track any human motion
  • OpenArm: 7DOF humanoid arm for physical AI research in contact-rich environments

Videos

  • Shared autonomy for interactive robotics: The Royal Society
  • Shared Autonomy: the robots are ready, are you? | Sethu Vijayakumar | TEDxUniversityofEdinburgh: TEDx Talks

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