Dekang Qi, Shuang Zeng, Xinyuan Chang, Feng Xiong, Shichao Xie, Xiaolong Wu, Mu Xu
MerNav introduces a Memory-Execute-Review framework that significantly improves success rates and generalization in zero-shot object goal navigation tasks.
Researchers have developed a new approach called MerNav to help robots navigate and find objects more effectively without prior training. This method uses a three-part system: a memory module to store information, an execution module to make routine decisions, and a review module to handle unexpected situations. When tested on various datasets, MerNav showed notable improvements in finding objects successfully, outperforming existing methods that either require training or don't generalize well.
