Socially Adaptive Path Planning in Human Environments Using Inverse Reinforcement Learning

A key skill for mobile robots is the ability to navigate efficiently through their environment. In the case of social or assistive robots, this involves navigating through human crowds. Typical performance criteria, such as reaching the goal using the shortest path, are not appropriate in such environments, where it is more important for the robot to move in a socially adaptive manner such as respecting comfort zones of the pedestrians. We propose a framework for socially adaptive path planning in dynamic environments, by generating human-like path trajectory. Our framework consists of three modules: a feature extraction module, inverse reinforcement learning (IRL) module, and a path planning module. The feature extraction module extracts features necessary to characterize the state information, such as density and velocity of surrounding obstacles, from a RGB-depth sensor. The inverse reinforcement learning module uses a set of demonstration trajectories generated by an expert to learn the expert's behaviour when faced with different state features, and represent it as a cost function that respects social variables. Finally, the planning module integrates a three-layer architecture, where a global path is optimized according to a classical shortest-path objective using a global map known a priori, a local path is planned over a shorter distance using the features extracted from a RGB-D sensor and the cost function inferred from IRL module, and a low-level system handles avoidance of immediate obstacles. We evaluate our approach by deploying it on a real robotic wheelchair platform in various scenarios, and comparing the robot trajectories to human trajectories.