Deep reinforcement learning for map-less goal-driven robot navigation

Problem

Mobile robots normally need an explicit map of the environment to navigate, which is impractical or impossible in new or dynamic environments. The study developed a local navigation method that steers a robot to global goals without a map.

Why it matters

Safe map-less navigation is crucial for robot autonomy in unknown or changing real-world environments.

What the AI did

Trained a deep reinforcement learning navigation model based on the Advantage Actor-Critic method that directly translates robot observations into movement commands, and compared it with standard map-based approaches in several simulated navigation scenarios.

Results

In simulation, the method navigated without a map or with a corrupted map while standard map-based approaches failed; no numeric metrics stated in abstract.

Impact

Demonstrated in simulation a navigation capability that survives map loss or corruption where standard approaches fail.

Record

Importance rating

More in Robotics & Autonomy