Event
AgriVLN: A Benchmark and Method for Vision-and-Language Navigation in Agricultural Robots
1 reports · 1 sources
The research team led by Xiaobei Zhao proposed A2A, a vision-and-language navigation benchmark for agricultural robots, along with a companion method, AgriVLN. A2A contains 1,560 episodes across 6 categories of agricultural scenarios and uses step-by-step instructions and a continuous-action environment. Evaluation found that existing vision-and-language navigation methods perform relatively well on short instructions and relatively poorly on long instructions. To address this, the team proposed an STL module that decomposes complex instructions into lists of simple subtasks, and integrated it into the NavGPT backbone to form the AgriVLN method. The authors report that this method raised the success rate (SR) from 0.33 to 0.47, while the navigation error (NE) increased slightly from 2.76 meters to 2.91 meters. The study also includes ablation experiments and case analyses, discussing the method's effectiveness and limitations under different numbers of subtasks and scenario categories.
Report timeline
- AgriVLN: A Vision-and-Language Navigation Benchmark and Method for Agricultural Robots
This work proposes A2A, a vision-and-language navigation benchmark for agricultural robots, comprising 1,560 episodes across 6 classes of agricultural scenes, using step-by-step instructions and a continuous-action environment. Evaluation found that existing VLN methods perform well on short instructions but poorly on long instructions; to address this, the authors propose an STL module that decomposes complex instructions into lists of simple subtasks and integrates it into the NavGPT backbone, raising SR from 0.33 to 0.47 while NE slightly increases from 2.76 m to 2.91 m. Ablation studies and case analyses are also included.
arXiv cs.RO ·