Reliability-Aware Future Conditioning for Temporally Robust

This research introduces RAFC, a method that improves the reliability of future predictions in robot manipulation by treating temporal alignment as a control problem. RAFC outperforms existing approaches under temporal misalignment, achieving higher success rates on both simulated and real-world tasks.

This research introduces SimpleICL, a framework for in-context robot learning (ICL) that enables robots to infer and execute manipulation tasks from visual demonstrations. The paper defines a clear problem scope for ICL, addresses ambiguity in prompt interpretation, and demonstrates strong performance in both simulation and real-world settings without requiring massive pre-training or specialized data infrastructure. The authors fully open-source their data and training pipeline to facilitate systematic and reproducible research on robot ICL.