Deformable Sim-to-Real, LLM Benchmarks, and Explainable Planning

Sim-to-Real for Deformables via Graph Deformable Transformer

Tackles sim-to-real domain transfer for complex deformable objects. Useful for anyone trying to get a robot to handle soft items reliably without real-world data collection.

RoboBench: Simulating Diverse Environments for LLM Robotics

Introduces a benchmark for evaluating LLM-controlled robots in diverse, hazardous environments. Highlights current limitations and performance gaps.

Explainable MCTS for Robot Planning

Adds an explainability framework to Monte Carlo Tree Search for robot planning. Aims to make complex planning decisions in manipulation more interpretable.

RobotBEV: A Benchmark for Generalizable Robot Policies

A new benchmark for evaluating how well robot control policies and designs can adapt to novel tasks and morphologies, focusing on generalization.