Vision-based navigation with language-based assistance

Vision-based navigation with language-based assistance is a task that combines visual perception and language guidance, aiming to navigate an agent to find specific objects in realistic indoor environments through high-level language goals. This task simulates real-world scenarios where the requester only provides high-level objectives, and the agent can actively query a more experienced advisor for concrete language subgoals when it gets lost, thereby continuing the navigation process. This technology holds significant application value in enhancing the autonomy and interactive capabilities of robots.