Tao Wang

Language instructions directly shape latent traversability representations, enabling adaptive visual navigation that is $6\times$ faster than explicit segmentation-based approaches.

Velocity penalties in RL reward functions are not just redundant for stabilization—they actively induce brittle policy gradient training as their coefficients scale.

Standard text-prompt evaluations severely underestimate diffusion safety risks, which reliably resurface under embedding-level control and trigger policy enforcement failures unless probabilities are explicitly calibrated.

A novel air-ground collaboration method achieves a 77% joint success rate, showcasing the power of shared spatial context in navigation tasks.