DRL-SFM: Learning Social Navigation from Costmaps and Social Forces for Mobile Robots and Intelligent Wheelchairs
Matthias Kalenberg, Kilian Gerhard Probst, Andreas Gründer, Christopher May, Jonas Walter, Jörg Franke
AI summary
Problem
Existing DRL-based social navigation planners often learn overly passive, deferential behaviors that fail to capture the mutual adaptation and reciprocal negotiation characteristic of human-human interactions, which is especially problematic for passenger-transport platforms like intelligent wheelchairs.
Approach
The authors propose a DRL local planner that augments the standard Nav2 costmap with pedestrian velocity data (HuMap) and shapes its reward function using the Social Force Model to encourage forward-looking, reciprocal navigation policies.
Key results
- Higher success rate in crowded scenarios
- Fewer space intrusions and socially compliant trajectories
- Up to 11% performance gain over velocity obstacle-based DRL planners
- Validated in simulation and real-world intelligent wheelchair experiments
Why it matters
It provides a deployable, costmap-compatible framework that improves safe and comfortable social navigation for assistive robots and intelligent wheelchairs in real-world crowded environments.
Abstract
The demand for assistive robots for passenger transport, such as intelligent wheelchairs, is increasing rapidly due to demographic changes. To allow passengers to navigate in crowded environments, such as shopping malls and hospitals, these systems must navigate in a socially accepted manner that ensures the comfort of both passengers and surrounding pedestrians. Although deep reinforcement learning (DRL) has shown promising results for social navigation, existing planners often learn overly passive behaviors, not engaging in the mutual adaptation characteristic of human interaction. In this paper, we introduce a novel DRL-based local planner that learns nav- igation behaviors by integrating the Social Force Model (SFM) directly into its reward function, allowing more cooperative interactions for mobile robots and intelligent wheelchairs. This approach encourages the agent to learn more forward-looking and reciprocal navigation policies by rewarding actions that align with the dynamics of pedestrians. To ensure generaliza- tion and straightforward deployment, our method utilizes the standard Navigation 2 local costmap augmented with pedestrian detections as an observation. The experiments demonstrate that our agent achieves a higher success rate in crowded scenarios with fewer space intrusions, outperforming the state-of-the-art DRL planner based on velocity obstacles by up to 11 %.