Sensorless Force Estimation

Estimating the external force on a robot from motor current and a dynamics model, without installing a dedicated force sensor.

Sensorless force estimation infers the force a robot experiences during contact with its environment without installing a dedicated six-axis force/torque sensor or joint torque sensor. Instead, it uses motor current — multiplied by the torque constant to approximate the motor’s output torque — together with joint position and velocity and a dynamics model of the robot. The idea is that the model predicts how much torque would be needed with no external force, and the difference between the torque actually measured and that prediction is attributed to the external force. A commonly used tool is the generalized momentum observer, which avoids differentiating acceleration and so is less noisy. The benefit is skipping an expensive, fragile force sensor, letting low-cost arms, quadrupeds, and humanoids do collision detection and rough force control; the difficulty is that gearbox friction, model error, and temperature changes introduce significant error, so precision is generally worse than with a dedicated sensor. Recent work uses learned methods to compensate for the residual torque, such as K-VARK (2025) and a 2026 hybrid estimation method for wheel-legged robots.

ExampleAn arm with no force sensor compares its motor current against what the dynamics model predicts; a sudden large mismatch in joint torque is taken to mean it has hit a person, and it stops immediately.

Also called

Current-Based Force Estimation, Motor-Current Force Estimation

Related

Generalized Momentum Observer · Torque Constant (Kt) · Collision Detection (Robot Safety) · Friction Compensation · Proprioceptive Actuator · Six-Axis Force/Torque Sensor

Sources

arXiv 2512.13009: K-VARK for Sensorless Force Estimation in Collaborative Robots
arXiv 2609.13779: Force-Aware RL with Hybrid Sensorless Force Estimation for Wheeled-Legged Loco-Manipulation

As of

2026-09