Han Zhang @hanzhang.bsky.social · 28/09/2026Then, we built this model in an artificial agent, and evaluated its performance in a reach-avoid task: the agent must avoid obstacles to reach the goal: 100 0
Han Zhang @hanzhang.bsky.social · 28/09/2026Compared to MLP and transformers, the priority-based agent trained more efficiently and handled unseen, hazardous scenarios better, even from simple demonstrations. Below is the agent handling x4 speed at ease, even without being trained on it. 100 0
Han Zhang @hanzhang.bsky.social · 28/09/2026We could even produce a human-like, history-driven effect, where the agent anticipates the upcoming target location: 110 0
