A major component in MIT’s DARPA Robotics Challenge walking system for our copy of Boston Dynamic’s Atlas robot was the Pronto State Estimator. Our fantastic controller was mainly developed by Scott Kuindersma. The primary input to the controller is the state of the robot estimated using its sensors – position and velocity in 12 dimensions. Below are some details of the state estimator – called Pronto – developed for this purpose:
Continuous Localization enabled by inertial, kinematic and LIDAR sensing
By continuously incorporating position corrections from a LIDAR map, the position of the robot can be corrected while walking – allowing the robot to traverse this uneven terrain without stopping – and can even do so in reverse (without any rear facing sensors).
Random arm motions (l), compliance during disturbances (r):
Stability when carrying unmodeled masses (l), support for a flight phase (toward running and jumping (r):
The Pronto State Estimator is described in this published papers:
- Maurice F. Fallon, Matthew Antone, Nicholas Roy, and Seth Teller. Drift-free hu- manoid state estimation fusing kinematic, inertial and lidar sensing. Humanoids, 2014.
- S. Kuindersma and R. Deits and M. Fallon and A. Valenzuela and H. Dai and F. Permenter and T. Koolen and P. Marion and R. Tedrake, Optimization-based locomotion planning, estimation, and control design for Atlas, JAR 2015.
- The Pronto estimator is now open source. You can download it from here.