FLTRNN:Faithful Long-Horizon Task Planning for Robotics with Large Language Models

Environment Dependency

The experimental environment is virtualhome, and the virtualhome version of our experiment is 2.2.5. For the installation and configuration of virtualhome, please refer to the official documentation: http://virtual-home.org/tools/explore.html https://github.com/xavierpuigf/virtualhome

Run

Run the program by executing sh scripts/inference.sh in the behavior_cloning directory.

control flags:

if_gpt = True   # use the gpt api to perform task planning
if_exe_all_action = True  #execute the complete plan after the generation is completed, or to execute while generating.

Before execution, you need to add your own api-key in gpt_policy.py

api_key = [api-key1, api-key2]
api_key_num = 2

Related code directory

inference.sh： some control parameters
gpt_policy.py： Interaction interface files related to GPT
interactive_interface_fn： main process function
/checkpoint/LID-Text/interactive_eval： log files