Navigation with Large Language Models: Semantic Guesswork as a Heuristic for Planning: Prompts

18 Apr 2024

This is paper is available on arxiv under CC 4.0 DEED license.


(1) Dhruv Shah, UC Berkeley and he contributed equally;

(2) Michael Equi, UC Berkeley and he contributed equally;

(3) Blazej Osinski, University of Warsaw;

(4) Fei Xia, Google DeepMind;

(5) Brian Ichter, Google DeepMind;

(6) Sergey Levine, UC Berkeley and Google DeepMind.

B Prompts

B.1 Positive Prompt

Figure 7: LFG in an unseen apartment. The robot starts in the same starting location and environment as 5, and is tasked with finding an oven. LFG guides the robot towards the kitchen appliances, rather than the bedroom door, and successfully leads to the oven.

Figure 8: LFG in an unseen office building. The agent looks for a sink in an open-plan office building. Despite erroneous detections, the robot continues exploring the environment, with LFG guiding it towards frontiers containing appliances found in a cafe. The robot successfully finds the sink despite imperfect detections.

B.2 Negative Prompt