Saransh Gupta

I'm developing BEV world-model-based trajectory evaluation systems for autonomous driving at BMW TechWorks India, including future-state prediction, online trajectory scoring, and uncertainty quantification using learned representations and Gaussian Mixture Models.

Previously at Flux Auto, I built a learned-dynamics simulator from forklift telemetry and deployed deep RL motion-planning for autonomous logistics vehicles. I also customized multi-agent RL for decentralized path planning and created a real-time GPS-failure recovery planner in C++.

My work spans sim-to-real transfer, planning, learned dynamics, production data pipelines, and reinforcement learning. Earlier at Tata Consultancy Services, I built a multi-armed bandit ad-selection system using real Wi-Fi telemetry.