Chrono::Sensor is a specialized module in Project Chrono for the modeling and simulation of sensors within a Chrono simulation. This simulation module is in development with current support for simulation of camera, lidar, GPS, and IMU. These sensors also include models for data generation, distortion, and noise. We are using ray tracing, high performance computing, and machine learning to develop and implement models and methods that can faithfully generate synthetic sensor data from within an evolving virtual environment managed by Chrono. The goal of this research is to enable software-in-the-loop training and evaluation of autonomous vehicles and robots. This research focuses on physics-based and machine-learned approaches for improved sensor realism, leveraging HPC for efficient sensor simulation, and methods for improvement of simulation-to-reality transfer. In addition, we are using this capability to train reinforcement learning algorithms for controlling vehicles and robots in on and off-road applications.
Contributors: Asher Elmquist, Han Wang, Aaron Young, and Eric Brandt