Robust Optimization-Based Control and Planning for Legged Robots Workshop

During the last 10 years the field of legged robots has been strongly influenced by the advent of efficient optimization techniques, which coupled with cheap and fast computers have allowed for the resolution of optimization problems inside high-frequency control loop. However, despite these recent advances, the results of the Darpa Robotics Challenge finals (June 2015) have clearly shown the lack of robustness of these control/planning algorithms: unmodeled uncertainties have often been the cause of failures.

This workshop aims to understand whether robust optimization could be the right tool to account for the countless uncertainties affecting legged robots, such as modeling errors, actuation inaccuracies, estimation uncertainties and delays. Bringing together people working on robust optimization, robust Model Predictive Control, and legged robots, we will try to answer to questions such as:

• What limits us in transferring from simulations to real robots? Is it modeling assumptions / bandwidth / uncertainty?
• Is robust optimization fast enough for application in control?
• Can improved robustness outweigh slower control rates?
• How to model and identify uncertainties?
• Which uncertainties are the most important to take into account?
• Are there modeling assumptions that make computation hard, but are not so important on real systems?