TOCALib: Optimal Control Library with Interpolation for Bimanual Manipulation and Obstacles Avoidance
Yulia Danik, Dmitry Makarov, Aleksandra Arkhipova, Sergei Davidenko, Aleksandr Panov
AI summary
Problem
Real-time optimal control for bimanual robots is computationally expensive and highly sensitive to initialization, while existing methods struggle to efficiently generate collision-free, dynamically feasible trajectories for complex tasks or reinforcement learning.
Approach
The authors precompute optimal trajectories using nonlinear programming with symbolic dynamics and differentiable collision detection, then approximate solutions for arbitrary endpoints via trilinear interpolation and splines.
Key results
- Symbolic optimization framework integrating kinodynamic constraints with differentiable collision detection
- Precomputed trajectory library with trilinear interpolation for fast endpoint approximation
- Low simulation-to-reality tracking error on the Mobile Aloha robot (0.0258 rad joint, 0.0170 m end-effector)
- Modular URDF handling and scalable self-collision constraint generation for multi-link manipulators
Why it matters
Accelerates real-time bimanual control and provides reliable, high-fidelity datasets for training reinforcement learning agents in complex manipulation tasks.
Abstract
The paper presents a new approach for construct- ing a library of optimal trajectories for two robotic manip- ulators, Two-Arm Optimal Control and Avoidance Library (TOCALib)1. The optimization takes into account kinodynamic and other constraints within the FROST framework. The novelty of the method lies in the consideration of collisions using the DCOL method, which allows obtaining symbolic expressions for assessing the presence of collisions and using them in gradient-based optimization control methods. The proposed approach is applicable for complex bimanual manipulations that require precision. In this paper we tested TOCALib on Mobile Aloha robot, as an example. The approach can be extended to other bimanual robots, as well as to gait control of bipedal robots. It can also be used to construct training data for machine learning tasks for manipulation.