15 Nov 2021
CoRL 2021 in Person
Last week members of ORI went down to London to the first in-person conference in nearly two years. The 5th annual Conference on Robot Learning (CoRL) was held between November 8 and 11 and was held in a hybrid format with about half of attendees online and half in-person in London.
At the conference, Siddhant Gangapurwala provided a demo of his work RLOC: Terrain-Aware Legged Locomotion using Reinforcement Learning and Optimal Control. This work implements a reinforcement learning-based controller for robust locomotion over mobility challenges on our ANYmal C quadruped robot. We showed how the robot is able to robustly walk of steps and inclines while following a joystick input from the user.
Russell Buchanan gave a poster presentation of his paper Learning Inertial Odometry for Dynamic Legged Robot State Estimation. The paper introduces a novel proprioceptive state estimator for legged robots based on a learned displacement measurement from IMU data. The learned inertial displacement measurement can improve state estimation in challenging scenarios where leg odometry is unreliable, such as slippery and compressible terrains.
The conference was a great success and it was a great relief to be meeting in-person again after so long. The strong in-person component also improved the online experience. Virtual attendees were able to join via Gather Town and interact with both the presenters and attendees in London. Posters were replaced with computer monitors with screens shared online to Gather Town. The conference ended with a banquet in the National Gallery.
Photos by Frank Fu