Large-Scale Robot Policy Evaluation with NVIDIA Isaac Lab-Arena | Robotics Office Hours
In this OpenUSD Insiders Robotics Office Hours session, we explore large-scale robot policy evaluation using Isaac Lab Arena, an open-source framework built for scalable task curation and GPU-accelerated benchmarking in simulation. You will see how to move from a single simulated task to thousands of parallel environments, and how to evaluate foundation models and robot policies at scale. This session covers: - Modular, Lego-like task composition - Swappable robots, objects, and scenes with minimal code changes - Affordance-based object interactions such as openable and pressable - Homogeneous and heterogeneous parallel GPU evaluation - Interoperability with teleoperation, imitation learning, RL, and dataset generation workflows - Integration with environment hubs and community benchmarks Isaac Lab Arena GitHub: https://github.com/isaac-sim/IsaacLab-Arena ---------------------------------------------------------------------------- Key Moments: 0:00 Intro and Robotics Office Hours kickoff 0:44 What is Isaac Lab Arena? 4:31 The scalability challenge in policy evaluation 5:51 Unified framework for task setup and evaluation 6:25 Modular Lego-like task composition 8:53 Task diversification and affordances 9:37 Large-scale GPU-parallel evaluation 10:51 Community benchmarks and deployment 12:22 Roadmap: v0.2 and beyond 15:28 Hands-on: Arena repository and setup 18:02 Building environments from modular components 23:06 Affordances: openable and pressable objects 25:46 Defining task success, reset, and metrics 28:34 Composing robots, scenes, and tasks 31:42 Swapping robots, objects, and scenes 34:43 Data generation and learning interoperability 37:43 Parallel evaluation demo (homogeneous) 41:36 Heterogeneous evaluation preview 42:45 Lightwheel large-scale benchmark results 47:57 Robo Finals and industrial benchmarks 48:38 Deformable object evaluation 50:52 Wrap-up and open source call to action 1:03:33 v0.2 preview: natural language placement and long-horizon tasks 1:10:07 GTC announcements and training sessions ---------------------------------------------------------------------------- 📆 Check out the full calendar for all of our upcoming events → https://nvda.ws/3JqaWnA ---------------------------------------------------------------------------- 👥Join the Community → https://nvda.ws/3ZMfc6e #IsaacLab #Robotics #EmbodiedAI #RobotLearning #PolicyEvaluation #Simulation #GPUComputing #FoundationModels #Manipulation #ReinforcementLearning #ImitationLearning #Lightwheel #OpenSourceRobotics #NVIDIAGTC