Push, Press, Slide: Mode-Aware Planar Contact Manipulation via Reduced-Order Models
New method replaces complex physics calculations with a library of intuitive models for faster robotic control.
A team of researchers has introduced a novel framework for planar robotic manipulation that sidesteps traditional computational bottlenecks. The paper, 'Push, Press, Slide: Mode-Aware Planar Contact Manipulation via Reduced-Order Models' by Melih Özcan, Ozgur S. Oguz, and Umut Orguner, tackles the notoriously difficult problem of non-prehensile manipulation—tasks like pushing or sliding objects without gripping them. These actions are critical in warehouses and factories but are challenging due to complex, hybrid contact mechanics and friction limits that usually require slow, iterative optimization to solve.
The core innovation is a 'mode-aware' approach that maps various contact scenarios between a robot arm and an object to a pre-defined library of simple kinematic models. For instance, a press-and-slide motion can be represented by a straightforward unicycle model. By anchoring planning to these reduced-order models, the framework can algebraically compute the required object forces and distribute feasible contact wrenches directly, without running heavy optimization solvers. This method was validated in simulation across diverse single-arm and bimanual tasks, showing it can ensure long-horizon feasibility while being significantly faster than conventional methods. The work, submitted to IEEE IROS 2026, represents a shift towards more efficient and intuitive control strategies for real-world robotic manipulation.
- Replaces complex iterative optimization with a library of simple kinematic models (e.g., unicycle) for fast planning.
- Uses a direct algebraic allocator to compute and distribute friction-bounded contact forces without optimization solvers.
- Demonstrated in simulation for diverse single-arm and bimanual non-prehensile manipulation tasks like pushing and sliding.
Why It Matters
Enables faster, more reliable robotic manipulation in logistics and manufacturing, moving complex objects without traditional computational overhead.