Symbolic control ensures safe robot motion from learned demonstrations
Learning from Demonstrations meets Symbolic Control - A Supervisory Control Architecture for Formal Certification of Learned Motion
Robotics
Summary
Robots can learn to move by watching examples, but it’s hard to guarantee they’ll always stay safe or reach their goals. The authors created a control system that watches the robot’s movements and steps in only when safety or success is at risk, without relying on the learned behavior itself. This system uses a backup plan based on the robot’s physics to make sure it never breaks important rules. They tested this approach on real and simulated robots doing tasks like handwriting, showing it helps keep robots safe and reliable even in tricky situations.
What this means in practice
- •For robotics engineers: Implement safe backup plans that automatically intervene in learned robot motions to ensure safety and goal completion.
- •For industrial automation teams: Deploy robotic systems with formal safety checks for complex tasks in unpredictable environments using learned motion models.
- •For industrial robotic integrators: Develop certified robotic products that combine flexible learning-based motion with guaranteed safety interventions for customer deployments.$Commercial implications: Enables creation of certified robotic motion control products guaranteed to safely handle learned tasks for industrial buyers.
Authors
Ashok Kumar, Mahathi Anand, Angela Schoellig, Abdalla Swikir
Abstract
Dynamical system (DS)-based learning from demonstration generates flexible robot motions, but existing guarantees rely on certificate generation to obtain formal guarantees of safety and goal convergence. We propose a symbolic supervisory control architecture that automatically generates a fail-safe backup plan, certifying safety, reachability, and actuator feasibility directly from the robot's dynamics, independent of the learned DS. A nominal controller tracks the learned DS, while the supervisory controller monitors safety and target convergence, providing non-conservative intervention when the nominal controller violates necessary constraints. Experimental results include simulations and real robot deployment with various 2D and 3D datasets, including handwriting datasets, validating the effectiveness of our approach in providing formal guarantees with learned motion even in complex, dynamic environments.