H-MaP: An Iterative and Hybrid Sequential Manipulation Planner
Berk Cicek, Arda Sarp Yenicesu, Cankut Bora Tuncer, Kutay Demiray, Ozgur S. Oguz
AI summary
Problem
Current manipulation planners struggle with high configuration space dimensionality and susceptibility to local minima when executing complex sequential tasks that require dynamic contact mode switches in constrained environments.
Approach
The planner decouples the problem by first generating a feasible object trajectory via sampling, then identifying valid physical contact points using a hybrid of neural inference and geometric sampling, and finally optimizing the robot's motion to follow the object path while satisfying contact constraints.
Key results
- Hybrid sampling-optimization framework for constrained sequential manipulation
- Iterative waypoint and contact point decomposition that reduces configuration space dimensionality
- Learning-informed contact sampling methodology for efficient feasible contact generation
- Successful validation across seven diverse single-arm and bimanual tasks in simulation and real-robot experiments
Why it matters
Advances practical robotic dexterity by enabling reliable execution of complex, multi-step manipulation tasks like tool use and bimanual coordination in cluttered or constrained spaces.
Abstract
This paper introduces H-MaP, a hybrid sequential manipulation planner that addresses complex tasks requiring both sequential actions and dynamic contact mode switches. Our approach reduces configuration space dimensionality by decoupling object trajectory planning from manipulation plan- ning through object-based waypoint generation, informed contact sampling, and optimization-based motion planning. This archi- tecture enables handling of challenging scenarios involving tool use, auxiliary object manipulation, and bimanual coordination. Experimental results across seven diverse tasks demonstrate H-MaP’s superior performance compared to existing methods, particularly in highly constrained environments where traditional approaches fail due to local minima or scalability issues. The planner’s effectiveness is validated through both simulation and real-robot experiments. https://sites.google.com/view/h-map/