Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
University of Haifa
6
0
5
Uncertainty in dynamic obstacle evolution, not partial observability, dictates whether classical or learning-based motion planning prevails.
End-to-end DRL for AUVs can achieve near-optimal navigation while significantly reducing engineering complexity.
Plug-and-play Kalman filtering can rescue pre-trained MARL policies from communication delays and packet loss, boosting performance without retraining.
Multi-objective RL reveals that micro-robot swarms can learn to exploit pulsatile flows for efficient upstream navigation, exhibiting emergent behaviors like hydrodynamic throttling and cycle-synchronized ratcheting.