Event

BM2PC: A Bayesian Multimodal MPC Framework for Dynamic Manipulation of Deformable Linear Objects

1 reports · 1 sources

Youyuan Long, Gokhan Solak, and Arash Ajoudani propose BM2PC, a Bayesian multimodal MPC framework for dynamic manipulation of deformable linear objects (DLOs). The framework models multimodal planning as a set optimization problem that balances control quality and behavioral diversity, and maintains multiple candidate control modes through incremental updates. The authors evaluated rope tracking and obstacle avoidance in simulation and on hardware, and tested disturbance rejection in simulation. In trajectory tracking, the authors claim that BM2PC requires no task-specific training or policy fine-tuning, and that its generalization to unseen trajectories outperforms a learning-based baseline; real-world experiments reduced RMSE by 44.0% on the longest task. In obstacle avoidance, the authors claim performance comparable to existing multimodal MPC methods while reducing planning time by 29% to 66%.