Bridging the Sim-to-Real Gap with Bayesian Inference
This work addresses the sim-to-real gap for robotics applications, offering a method that improves data efficiency and performance in model-based reinforcement learning.
The paper tackles the problem of bridging the sim-to-real gap in robot dynamics by introducing SIM-FSVGD, which uses low-fidelity physical priors to regularize neural network training, resulting in accurate mean model estimation and precise uncertainty quantification. It demonstrates effectiveness on an RC racecar system, achieving a highly dynamic parking maneuver with drifting using less than half the data compared to the state of the art.
We present SIM-FSVGD for learning robot dynamics from data. As opposed to traditional methods, SIM-FSVGD leverages low-fidelity physical priors, e.g., in the form of simulators, to regularize the training of neural network models. While learning accurate dynamics already in the low data regime, SIM-FSVGD scales and excels also when more data is available. We empirically show that learning with implicit physical priors results in accurate mean model estimation as well as precise uncertainty quantification. We demonstrate the effectiveness of SIM-FSVGD in bridging the sim-to-real gap on a high-performance RC racecar system. Using model-based RL, we demonstrate a highly dynamic parking maneuver with drifting, using less than half the data compared to the state of the art.