Abhishek Goudar

Planning in massive digital twins with point-cloud data becomes significantly faster by connecting belief states using a $W_2$-metric adaptation of RRT*, reusing measurement information to avoid repeated propagation.

CP-Cert achieves certification of optimality in degenerative non-convex problems at speeds up to 1,000 times faster than traditional solvers.

In-flight reinforcement learning is now feasible, with Crazyflow training a drone recovery policy in just 0.38 seconds mid-air.

Achieve decimeter-level accuracy in decentralized multi-robot formation control using only onboard sensors and inter-robot range measurements, even without carefully designed trajectories.

Need to hit a specific trajectory estimation accuracy with your robot? This work reverse-engineers the sensor requirements, telling you the necessary sensor update rate or noise covariance.