The field of robotics and control is witnessing a significant shift towards addressing the long-standing issue of the Sim2Real gap, which refers to the performance drop of policies and controllers when transferred from simulated environments to real-world scenarios. Recent developments have focused on developing innovative frameworks and algorithms that can effectively bridge this gap, enabling seamless transition of control policies from simulation to reality. A key direction in this area is the use of bi-level reinforcement learning frameworks, which adapt simulator parameters based on real-world performance, and sample-based hybrid mode control approaches, which enable asymptotically optimal switching between different control modes. Another important aspect is the development of simulation-based reinforcement learning algorithms that can handle uncertain and varying system parameters, and the use of domain randomization, real-to-sim transfer, and sim-real co-training techniques to mitigate the reality gap. Noteworthy papers in this area include:

• A Generalization of Input-Output Linearization via Dynamic Switching Between Melds of Output Functions, which proposes a systematic framework for switching between different sets of outputs for nonlinear systems.
• Closing the Sim2Real Performance Gap in RL, which introduces a novel bi-level RL framework to adapt simulator parameters based on real-world performance.
• Sample-Based Hybrid Mode Control: Asymptotically Optimal Switching of Algorithmic and Non-Differentiable Control Modes, which derives a sample-based variation to efficiently search for optimal solutions in hybrid mode control problems.