BSON is a platform to train and evaluate the socially-aware navigation algorithms in realistic and diverse social environments, based on the Unity 3D game engine. It is designed to provide the following features:
-
Various pedestrian motion models: BSON utilizes social force, ORCA, and MAC-ID to generate diverse pedestrian behaviors. It ensures a fair evaluation by preventing socially-aware navigation algorithms specialized in a particular pedestrian model from receiving inflating scores.
-
Various sensors: BSON offers support for RGB-D cameras, 3D Lidar, 2D Lidar, and IMU, enabling users to evaluate their algorithms using sensor configurations consistent with those of their real robot.
-
Supporting ROS2 and gym: We provide a simple python API based on gym for training and evaluating learning-based algorithms. If you use the ROS2 branch, you can obtain sensor information and control the robot via ROS2 messages, just like implementing a real robot system.
-
Easy customization: Designing custom environments and pedestrian trajectories is easily accessible to users.
The task of BSON is point goal navigation, which aims to navigate the robot towards a given point goal in a crowded environment. In this work, we have used the Jackal robot platform, bur it is possible to utilize different robots using URDF Importer.