Trajectory Planning for Safe Dual Control with Active Exploration

A team of researchers from the University of Michigan has introduced a novel framework called Dual-gatekeeper that addresses a fundamental challenge in robotics: how to plan safe trajectories when a robot's model of the world is uncertain. Traditional robust planning methods are often overly conservative, considering worst-case scenarios and ignoring opportunities to reduce uncertainty during a mission. The dual control problem—actively learning while performing a task—has typically been addressed by adding exploration terms to cost functions, but without formal guarantees about when exploration is actually beneficial or safe.

Dual-gatekeeper provides a principled solution by integrating robust planning with active exploration under strict constraints. The key innovation is that exploration is only pursued when it provides a verifiable improvement without compromising safety or violating a predefined mission-level cost budget. This budget limits how much task performance can degrade due to exploration, ensuring the robot doesn't wander aimlessly. The researchers provide two implementations based on different safety mechanisms and demonstrate the framework's effectiveness through case studies in quadrotor navigation and autonomous car racing under parametric uncertainty, showing how systems can balance immediate performance with long-term learning.