Test and Evaluation of Quadrupedal Walking Gaits through Sim2Real Gap Quantification
This work addresses the challenge of verifying system performance in robotics, particularly for quadrupedal walking gaits, but it appears incremental as it builds on existing methods like Bayesian Optimization and Sim2Real gap quantification.
The authors tackled the problem of evaluating a system's ability to meet operational objectives by proposing a two-step optimization approach using Bayesian Optimization, which quantifies the Sim2Real gap between simulators and hardware, showing repeatability and the ability to discriminate between different environments.
In this letter, the authors propose a two-step approach to evaluate and verify a true system's capacity to satisfy its operational objective. Specifically, whenever the system objective has a quantifiable measure of satisfaction, i.e. a signal temporal logic specification, a barrier function, etc - the authors develop two separate optimization problems solvable via a Bayesian Optimization procedure detailed within. This dual approach has the added benefit of quantifying the Sim2Real Gap between a system simulator and its hardware counterpart. Our contributions are twofold. First, we show repeatability with respect to our outlined optimization procedure in solving these optimization problems. Second, we show that the same procedure can discriminate between different environments by identifying the Sim2Real Gap between a simulator and its hardware counterpart operating in different environments.