Robotics core · topic B8 · 35 hours
Force control & impedance control
Position control can only say «where»; force control says «how hard». Assembly, polishing, human-robot collaboration, impact resistance — all of it lives here.
What to learn
- Impedance versus admittance control: one commands position and yields in force, the other commands force and yields in position — who adapts to whom
- Stiffness, damping, and stability: why more latency forces a lower achievable stiffness
- Force-sensor calibration, gravity compensation, zero-drift handling
- Sensorless collision detection: the current-residual and momentum-observer methods
The lesson · plain words
Position control answers «arrived or not»; force control answers «pressing how hard». Driving screws, peg insertion, wiping glass, shaking hands — wherever the robot and an object touch, it is this page's business.
The single core concept is impedance control: it turns the arm's tip into a spring whose hardness is a knob.
What happens when position control meets a wall
Position control's essence: «the bigger the position error, the harder the push». Touch a wall, the error never resolves, the controller keeps pushing harder — the harder the wall, the bigger the force, and there is no ceiling.
On hardware: motors hum against the deadlock, current alarms trip, the workpiece cracks. This is not bad tuning — the position-control framework simply has no answer for contact.
Impedance control: stiffness becomes a knob
Flip the thinking: stop chasing position, and instead command «tip force = stiffness × position error − damping × velocity». The tip behaves like a spring whose stiffness K you set.
Drag the stiffness slider below and watch the peak contact force fall from hundreds of newtons to tens. That is why collaborative arms dare to share a room with people.
Interactive figure · drag it and see
Two implementations: direct force control versus impedance
Direct force control: mount a six-axis force-torque sensor, close the loop on force error directly. Fast and precise — and expensive, noisy, and prone to vibrating on compliant structures.
Impedance control: a stiffness setting wrapped around a position inner loop; it works without any force sensor (torque estimated from motor current). The overwhelming majority of industrial «force control» is actually impedance.
Every hard part lives at contact
Contact stiffness is unknown: push a sponge or push a steel plate — the same K behaves completely differently. Latency: a few tens of milliseconds of delay oscillate outright at high stiffness. Friction: tangential forces turn «wipe the glass» into a trajectory-plus-force coupling problem.
The classic resolution is hybrid control: decompose the task into a force-controlled direction and position-controlled directions — screwing a fastener presses along the axis under force control while rotating under position control. Once you see the decomposition, many tasks become legible.
Three questions — do you really get it?
1. A position-controlled arm hits a rigid wall — what most likely happens?
- AIt decelerates and stops by itself
- BThe force grows without bound — deadlock or a crushed workpiece
- CIt switches to force-control mode
Show answer
B. The position error never resolves and the output saturates. The entire reason force control exists is to patch this hole.
2. In impedance control you lower the stiffness K — what happens on contact?
- AContact force grows
- BPeak contact force drops and the tip settles in earlier
- CPosition tracking becomes more accurate
Show answer
B. A small K produces less force for the same penetration. The price: tracking turns soft and free-space error grows.
3. How is most industrial «force control» actually implemented?
- AA position inner loop plus a stiffness setting (impedance)
- BDirect closed loop on a six-axis F/T sensor
- CHydraulics
Show answer
A. Impedance does not need a force sensor — cheaper, stable, and current-loop torque estimation is already good enough.
Keep going deeper
Links without a marker stay on this site; the rest point to the Chinese edition until its translation batch lands.
How it fails in the field
- Trying to tune stiffness «hard as a wall» — it oscillates on first touch; latency simply does not allow it
- Skipping thermal-drift compensation on the force sensor: after an hour the zero has wandered and the calibration was for nothing
You have learned enough when
The Chinese edition carries the full topic page for Force control & impedance control, including the complete outline, depth requirements per career track, and further derivations: robotguide.dev/k/B8/.