Benchmarking Structured Policies and Policy Optimization for Real-World Dexterous Object Manipulation

Dexterous manipulation is a challenging and important problem in robotics. While data-driven methods are promising approach, current benchmarks require simulation or extensive engineering support due to the sample inefficiency of popular methods. We present for TriFinger system, an open-source robotic platform dexterous focus 2020 Real Robot Challenge. The benchmarked methods, which were successful challenge, can be generally described as structured policies, they combine elements classical robotics modern policy optimization. This inclusion inductive biases facilitates efficiency, interpretability, reliability high performance. key aspects this benchmarking validation baselines across both real thorough ablation study over core features each solution, retrospective analysis challenge benchmark. code demo videos work found on our website (https://sites.google.com/view/benchmark-rrc).