Contact Estimation

Determines whether each foot of a legged robot is currently on the ground or in the air.

Contact estimation is a subproblem of state estimation for legged robots: at every control cycle, it decides whether each leg is touching the ground, usually outputting a contact probability. There are several general approaches: reading foot-force sensors and thresholding them; using joint torques and a dynamics model to infer ground reaction force; probabilistically fusing gait phase and foot height; or, as in Lin and colleagues (CoRL 2021), training a neural network to learn touchdown events directly from proprioceptive data like joint encoders and the IMU (inertial measurement unit), with no dedicated contact sensor needed. Downstream modules all depend on this judgment: leg odometry assumes a foot in contact isn’t moving and uses that to infer body velocity, so a wrong call causes drift; the controller also relies on it to switch between stance and swing phases.

ExampleIf a quadruped robot trotting over gravel misjudges a slipping foot as “stably in contact,” leg odometry built on an invariant extended Kalman filter will mistake the slip for body motion, and the position estimate will drift as a result.

Also called

Touchdown Detection, Foot Contact Estimation

Related

Leg Odometry · Invariant Extended Kalman Filter · State Estimation · Foot Force Sensor · Learned State Estimator · Stance Phase / Swing Phase

Sources

Legged Robot State Estimation using Invariant Kalman Filtering and Learned Contact Events (arXiv 2106.15713, CoRL 2021)