Safe Optimal Control Framework for Cooperative Manipulation of Objects in Human-Robot Teams

Area of Science:

• Robotics
• Artificial Intelligence
• Control Theory

Background:

• Cooperative object manipulation in human-robot teams presents challenges due to unknown agent dynamics and the need for real-time intent estimation.
• Existing frameworks often lack distributed estimation capabilities and robust safety mechanisms for complex multi-agent interactions.

Purpose of the Study:

• To introduce a distributed deep neural network (NN)-based adaptive control framework for cooperative object manipulation in human-robot teams.
• To enable accurate estimation of human intent and robust control of robotic agents with unknown dynamics.
• To ensure safety and efficiency in multi-agent coordination through advanced estimation and control strategies.

Main Methods:

• Utilized three distinct multilayer NN observers (MNNOs) for reference point estimation, human force-to-trajectory estimation, and distributed NN dynamics observation.
• Employed consensus-based learning for distributed state estimation without global trajectory access.
• Integrated a distributed online actor-critic NN controller with Pareto game theory and barrier Lyapunov functions (BLFs) for optimal, safe control.
• Developed weight update laws using singular value decomposition (SVD) for stable parameter tuning.

Main Results:

• Achieved accurate real-time estimation of human intent (position, velocity, acceleration) through force-to-trajectory inference.
• Demonstrated robust control performance in cooperative object manipulation tasks with unknown agent dynamics.
• Reported a significant 60% reduction in total cost compared to baseline methods.
• Validated the effectiveness of distributed estimation and NN-based adaptive control in multi-agent settings.

Conclusions:

• The proposed framework effectively addresses challenges in human-robot team coordination for object manipulation.
• The integration of multiple NN observers and an actor-critic controller ensures accurate intent recognition and adaptive, safe control.
• The framework offers a promising solution for enhancing collaboration and efficiency in human-robot systems.