Impedance Control: Planning or Learning? by Antonio Bicchi
Antonio Bicchi, professor at the University of Pisa and researcher at the Italian Institute of Technology in Genoa, discusses planning and learning robot impedance, with examples from his group and collaborators.
In this talk
- 0:49
Why compliance matters
Bicchi describes compliance as a source of dexterity, energy efficiency, safety, and resilience in animals and robots. In people, muscle forces and co-contraction affect joint stiffness. He distinguishes posture-driven changes to stiffness direction from co-contraction, which mainly changes its magnitude.
- 7:46
Mechanical and software approaches
Robots can vary impedance through software control or through actuators whose physical stiffness changes. Bicchi gives examples including a modular variable-stiffness actuator, an exoskeleton that uses leg muscle signals to assist standing, and a prosthetic elbow designed to soften impacts. Software impedance control is configurable and precise, while physical compliance can offer greater resilience and energy efficiency.
- 16:46
Planning impedance for the task
Bicchi outlines optimal planning, human control, and learning from people as ways to use variable impedance. Optimization examples show that the useful stiffness depends on the goal: safer fast motion calls for softness at high speed, while a hammer swing benefits from stiffness during acceleration and softness during deceleration. For contact tasks, choosing the robot’s posture and impedance can help avoid instability or excessive force.
- 26:43
Human control and learning
In tele-impedance, a person controls both a robot’s position and stiffness, with muscle activity influencing how forcefully it interacts. Bicchi describes applications in peg insertion and prosthetic hands, including resisting disturbances through co-contraction. In a teaching example, recording human stiffness alongside motion lets a robot switch between firm positioning and softer contact, helping it place a peg successfully even when conditions vary.
- 41:09
Takeaways for robot design
Human impedance changes both across tasks and during a task, through muscle co-activation and posture. Robots can vary impedance through software, hardware, and posture, and can use optimization or human input to select it. Bicchi concludes that a small number of demonstrations may be enough for robots to learn useful impedance behavior.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.