Hi NVIDIA Brev & Academic Support Team,
I’m a Master’s student in Robotics at the General Robotics, Automation, Sensing & Perception (GRASP) Lab at the University of Pennsylvania.
Simulating dense multi-agent aerial environments in parallel, where dozens of quadcopters interact in close quarters with continuous tensor updates for proximity sensing and collision bounds, quickly exhausts my local GPU memory. To train multi-agent RL (MARL) algorithms at large batch sizes, I need ray-tracing and PhysX performance beyond what my local hardware can provide.
Technical scope & workflow
- Institution: UPenn GRASP Lab
- Core stack: NVIDIA Isaac Sim 6.0, Isaac Lab, PyTorch (MARL / MAPPO workflows), Docker
- Research focus: Multi-drone close-quarters flight, inter-agent proximity dynamics, and tight-formation collision avoidance
- Hardware target: 1x NVIDIA L40S instance on Brev via Isaac Launchables
Community contribution
I plan to open-source the multi-drone close-quarters simulation environments and training scripts on GitHub so other robotics researchers can benchmark multi-agent aerial RL in Isaac Lab.
Any promotional GPU credits on Brev would directly support this work on NVIDIA infrastructure. Thank you for supporting academic robotics research!
Best regards,
Jaime Romero
M.S. Robotics, GRASP Lab | University of Pennsylvania
Website: https://jaromero3rd.github.io/
GitHub: jaromero3rd (Jaime Romero) · GitHub
LinkedIn: https://www.linkedin.com/in/jaime-romero-b7217a198/
Recent paper: Visual Navigation Transformer with Pose Attention, [2609.21212] Visual Navigation Transformer with Pose Attention