Robotics
Contact-Rich Manipulation
ActivePublished: 29 September 2026Updated: 29 September 2026Published
Key
innovation
Robotic manipulation whose success depends on actively managing contact forces, friction and stiffness — not just free-space trajectories.
Category
Robotics
Abstraction level
Pattern
Operation level
Robot controlInference
Use cases
Industrial assembly (peg-in-hole)Polishing and deburringFolding cloth and deformable objectsTool useHuman–robot collaboration
How it works
The robot measures forces/torques (wrist F/T sensor, tactile) and regulates interaction with force or impedance control laws instead of rigidly tracking position. Learned policies (behavior cloning, RL, DAgger) map vision+touch+state to actions; newer approaches inject reactive force signals into VLA policies to cope with contact uncertainty.
Problem solved
Position-only control fails on contact: a stiff robot driving to a target position generates unsafe forces and breaks parts. Forces and compliance must be regulated explicitly.
Components
Force/impedance controlSafe contact interaction
A control law regulating the force–position relationship instead of position alone.
Force/torque and tactile sensingFeedback
Measurement of contact forces (6-axis F/T, tactile skin).
Learned manipulation policyAction decision
A model mapping perception and touch to actions (IL/RL/VLA).