Learning Social Cost Functions for Human-Aware Path Planning

TL;DR

The paper tackles socially aware path planning by learning a social cost function that augments existing grid-based planners, enabling robots to respect social norms even when humans are stationary. It proposes an encoder–decoder neural network that maps a social grid map M_S to a social cost map C_S, integrated as an additive term in the planner’s total cost f_c. The method is validated first in simulation and then on a real TIAGo robot for two scenarios—queuing and interaction spaces of groups—demonstrating robust behavior with high success rates (>95%). The approach preserves traditional navigation properties while enabling nuanced social behaviors, and it offers a modular plug-in suitable for extension to additional social scenarios and norms.

Abstract

Achieving social acceptance is one of the main goals of Social Robotic Navigation. Despite this topic has received increasing interest in recent years, most of the research has focused on driving the robotic agent along obstacle-free trajectories, planning around estimates of future human motion to respect personal distances and optimize navigation. However, social interactions in everyday life are also dictated by norms that do not strictly depend on movement, such as when standing at the end of a queue rather than cutting it. In this paper, we propose a novel method to recognize common social scenarios and modify a traditional planner's cost function to adapt to them. This solution enables the robot to carry out different social navigation behaviors that would not arise otherwise, maintaining the robustness of traditional navigation. Our approach allows the robot to learn different social norms with a single learned model, rather than having different modules for each task. As a proof of concept, we consider the tasks of queuing and respect interaction spaces of groups of people talking to one another, but the method can be extended to other human activities that do not involve motion.

Figures (6)

• Figure 1: Representation of the pipeline used to integrate the social costs into the traditional navigation system cost function. The social grid map $M_S$ is obtained from the position of the goal and people. The planner will compute the shortest path towards the goal considering the social cost map $C_S$ in addition to all the other obstacle cost maps.
• Figure 2: Representation of the Encoder-Decoder neural network used in this work.
• Figure 3: Simulation of navigation in a demo environment with three different groups of people interacting with each other and one queue, visualized in $(a)$ with Gazebo. Magenta patches in the cost map represent areas with high costs corresponding to obstacles, while the cyan-to-blue gradient is a decreasing cost that inflates the obstacles. The green line is the computed path towards the goal, positioned at the end of the queue. In $(b)$, the social cost map is included, allowing the navigation algorithm to plan a path that drives the robot at the beginning of the queue without crossing any group of people. On the contrary, in $(c)$ the social cost map is not considered; the robot cuts the line directly to the goal and passes between a group of people talking.
• Figure 4: Simulation of the robot approaching the queue from different directions: from the front $(a)$, the side $(b)$, and the back $(c)$ of the line of people. As can be seen, with the social cost map, the planner is always able to set the robot in the queue correctly.
• Figure 5: Simulation of the robot moving between multiple groups of people $(a)$ and approaching a long queue of people $(b)$. The network can recognize and delimit the whole queue and each group, assigning high costs to the interaction areas between the people.