Leg Odometry

Uses joint encoders, the IMU, and contact detection to work out how far and in what direction a legged robot has moved.

Leg odometry is a method for a legged robot to estimate its body position and velocity using only its own sensors (proprioception). The basic idea: while a supporting leg’s foot is planted on the ground and not moving, joint encoder readings and forward kinematics give that foot’s position relative to the body, which can be used to infer how the body itself is moving; this is then fused with the IMU’s acceleration and angular velocity, usually with a Kalman-filter-family method. ETH’s Bloesch and colleagues proposed at RSS 2012 putting the foothold position itself into the extended Kalman filter’s state, requiring no assumptions about the terrain, and validated this on a quadruped; the later invariant extended Kalman filter converges more reliably on the Cassie biped. It is unaffected by lighting or texture and runs at high frequency, making it a foundational input for motion control; but foot slip or a wrong contact judgment introduces error, and it drifts over long durations, so it is often corrected with visual or lidar odometry.

ExampleMIT Cheetah 3 and Mini Cheetah’s open-source LinearKFPositionVelocityEstimator uses a linear Kalman filter to estimate body position and velocity, with the IMU as the prediction step and foot position and velocity computed from leg kinematics as the measurement.

Also called

Proprioceptive Odometry, Kinematic Odometry

Related

State Estimation · Proprioception · Contact Estimation · Extended Kalman Filter · Invariant Extended Kalman Filter · Wheel Odometry

Sources

State Estimation for Legged Robots - Consistent Fusion of Leg Kinematics and IMU (Bloesch et al., RSS 2012)
Contact-Aided Invariant Extended Kalman Filtering for Robot State Estimation (arXiv 1904.09251)
MIT Cheetah-Software: PositionVelocityEstimator.h