search

Ram81 / pirlnav

Code for training embodied agents using IL and RL finetuning at scale for ObjectNav
53 stars 11 forks source link

The results of released the models. #4

Open yusirhhh opened 7 months ago

yusirhhh commented 7 months ago

When I utilize the released behavior cloning model, the results are as follows:

41,030 Average episode reward: 0.6090
2024-03-15 22:14:41,030 Average episode distance_to_goal: 3.3111
2024-03-15 22:14:41,030 Average episode success: 0.6090
2024-03-15 22:14:41,030 Average episode spl: 0.2669
2024-03-15 22:14:41,030 Average episode softspl: 0.3066
2024-03-15 22:14:41,030 Average episode sparse_reward: 1.5225

It has decreased by 4 points compared to the result in the readme.md, which was 64.

Can you give me some suggestions?

Ram81 commented 5 months ago

Hi @yusirhhh , which version of HM3D Objectnav dataset are you using?