Stay up-to-date with a variety of topics ranging from daily news to career advice, as well as monthly insights on AI, sustainability, software, and more by subscribing to our newsletter. Receive expert perspectives, exclusive content, and in-depth coverage of engineering and innovation, all with a reduced number of ads or an ad-free experience.
The COMPASS system, developed by researchers, leverages NVIDIA’s X-Mobility policy with Reinforcement Learning (RL) to automate and optimize the training, testing, diagnosis, and evaluation of AI agents for robot navigation systems. This innovative framework aims to simplify the adaptation of navigation policies when faced with changes in robots, scenes, or operating conditions. By incorporating AI agents, simulation, reinforcement learning, and automated testing, COMPASS streamlines the development process, allowing for more efficient navigation training and evaluation, while involving human input at critical decision points.
Navigating a robot involves a complex set of tasks, from understanding its environment to planning a safe route while avoiding obstacles. When faced with changes in the robot or its surroundings, developers usually need to undergo a time-consuming process of retraining and testing. COMPASS addresses these challenges by utilizing a pretrained navigation model and adapting it through reinforcement learning for specific robots and environments. AI coding agents further automate tasks such as software checks, simulation setup, initial testing, training, and model evaluation, while developers retain control over key decision-making processes.
Developers can kickstart their navigation system development using the reference workflow with the Boston Dynamics Spot quadruped robot in a built-in warehouse environment. COMPASS also offers the flexibility to utilize scenes from NVIDIA’s SAGE-10K dataset or reconstruct real-world environments using NVIDIA Omniverse NuRec for more realistic simulations. The system employs a systematic approach starting with smoke tests, progressing to reinforcement learning runs, and continuously evaluating trained policies based on predefined performance metrics.
Ultimately, COMPASS aims to provide a more efficient and reproducible way to adapt navigation systems across diverse robots and environments. By combining advanced technologies with human oversight for critical decisions, this framework revolutionizes the development process for robot navigation systems.
Jijo, an automotive and business journalist based in India, holds a BA in History (Honors) from St. Stephen’s College, Delhi University, and a PG diploma in Journalism from the Indian Institute of Mass Communication, Delhi. With experience in news agencies, national newspapers, and automotive magazines, Jijo enjoys off-roading, engaging in political discussions, traveling, and teaching languages in his free time.
About the Author
