Advances in Bridging the Sim-to-Real Gap in Robotics

The field of robotics is moving towards addressing the significant challenge of transferring control and planning policies from simulation to real-world environments, known as the sim-to-real gap. Recent developments focus on innovative methods to reduce this gap, including the use of conformal mapping, sample-efficient reinforcement learning, and bootstrapped model predictive control. These approaches aim to improve the efficiency and effectiveness of robotic applications in real-world deployments. Noteworthy papers include: A Schwarz-Christoffel Mapping-based Framework for Sim-to-Real Transfer in Autonomous Robot Operations, which proposes a lightweight conformal mapping framework to transfer control and planning policies. Sample-Efficient Reinforcement Learning of Koopman eNMPC, which combines model-based RL with Koopman eNMPC to achieve superior control performance and higher sample efficiency. Bootstrapped Model Predictive Control, which introduces a novel algorithm that performs policy learning in a bootstrapped manner, leading to better value estimation and improved efficiency.

Sources

A Schwarz-Christoffel Mapping-based Framework for Sim-to-Real Transfer in Autonomous Robot Operations

Sample-Efficient Reinforcement Learning of Koopman eNMPC

Bootstrapped Model Predictive Control

Bridging the Sim-to-real Gap: A Control Framework for Imitation Learning of Model Predictive Control

Empirical Analysis of Sim-and-Real Cotraining Of Diffusion Policies For Planar Pushing from Pixels

Built with on top of