Amanda Prorok

Physical robot kinematics can double as an out-of-band data link, letting agents transmit arbitrary bitstreams purely through camera-observable action noise without degrading policy performance or touching an RF transceiver.

Separating world factors from task factors enables robots to generalize across diverse environments and tasks, achieving zero-shot performance and hardware transfer without retraining.

Optimizing the environment itself, alongside agent actions, unlocks safer and more efficient multi-agent navigation.

By strategically applying differential privacy in the wavelet domain, this work achieves a significantly better privacy-utility tradeoff for image generation, suggesting that not all image frequencies are created equal when it comes to privacy risk.