Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
National Taiwan University, Taiwan
2
0
4
0
GPT-4o can't handle commonsense planning with changing object affordances as well as a LoRA-finetuned vision-language model trained for the task.
LLMs can generate plausible navigation instructions that lead agents on wild goose chases, revealing critical gaps in current VLN agents' ability to handle infeasible goals.