New learning-based navigation framework helps robots safely navigate indoors
Combines neural global planner with PPO-refined local control for obstacle avoidance.
This paper presents a learning-based navigation framework for indoor mobile robots. It combines a supervised neural global planner trained from cost-aware A* expert trajectories with a Learning-Based DWA local planner, formulated as discrete candidate selection over the Dynamic Window Approach action lattice. The local policy is first trained via behavior cloning, then refined using Proximal Policy Optimization with feasibility-aware masking. Tested in simulated and real-world indoor environments, the system generates feasible global routes and reliable local motion commands for safe goal-directed navigation amidst obstacles. Source code will be released at the provided link.
- Global planner uses a supervised neural network trained on cost-aware A* expert trajectories for optimal path planning.
- Local planner (Learning-Based DWA) is trained via behavior cloning then refined with PPO and feasibility-aware masking for safe motion.
- Validated in both simulation and real-world indoor environments, achieving safe obstacle avoidance and reliable goal-directed navigation.
Why It Matters
Enables safer, more reliable autonomous navigation in warehouses, hospitals, and homes without expensive sensor suites.