Dzmitry Tsetserukou

Forget sparse, catastrophic rewards – GSDrive uses differentiable 3D Gaussian Splatting to provide dense, physics-based feedback, dramatically improving end-to-end driving policy learning.

Imagine robots that not only understand your instructions but also instantly adjust their movements to ensure your safety and comfort, all thanks to a VLM-RAG system that dynamically modulates MPC parameters.

Autonomous racing perception isn't just about speed; EagleVision reveals that motion distribution coverage trumps in-domain training for trajectory prediction, while real-world pre-training beats simulation for object detection.

Robots can now grasp fragile objects more reliably by "feeling" out the strongest points on their surface, thanks to a novel force-map-aware grasp planning system.

Tactile sensing in robotics can be effectively learned offline and distilled into VLAs, enabling contact-rich manipulation without requiring expensive and platform-specific tactile sensors at inference time.

Ditch the map: a diffusion model learns to plan UAV swarm trajectories directly from RGB images, enabling reactive and adaptive navigation in cluttered environments.

Adaptive SINDy lets drones fly smoother in turbulence by learning to predict and counteract wind gusts better than traditional controllers.

Marrying NASA's F' flight software with ROS2 yields a robust, real-time autonomous flight system validated by a 32-minute indoor quadrotor flight.

Ditch the maps and heavy planning pipelines: HumanDiffusion lets UAVs navigate directly to humans in need using only a camera and a lightweight diffusion model.