Reactive collision-free motion generation in joint space via dynamical systems and sampling-based MPC.

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Dynamical System (DS) based motion planning provides reactive, collision-free paths but struggles with non-convex obstacles and high-dimensional spaces.
• Sampling-based Model Predictive Control (MPC) generates collision-free paths but is computationally intensive and limited to quasi-reactive scenarios.

Purpose of the Study:

• To develop a robot motion planning approach that combines the strengths of DS and MPC for enhanced reactivity and obstacle avoidance in cluttered environments.
• To enable robots to navigate complex, high-dimensional joint spaces while avoiding both static and dynamic obstacles.

Main Methods:

• Modulating joint-space DS with obstacle-tangential velocity components derived from asynchronously generated MPC paths.
• Utilizing MPC only when local minima are detected, reducing computational load.
• Deflecting nominal DS with tangential velocity components to navigate around obstacles and escape local minima.

Main Results:

• The proposed approach successfully avoids concave obstacles and maintains local attractor stability.
• Demonstrated capability in both quasi-static and highly dynamic cluttered environments.
• Validated through simulations and real-world experiments on a 7-DoF robot.

Conclusions:

• The hybrid DS-MPC approach offers a robust solution for generating feasible, highly reactive, and collision-free robot motion in complex environments.
• This method overcomes limitations of traditional DS and MPC techniques, particularly in scenarios with non-convex obstacles and dynamic changes.