Correct values for collision and offroad rewards randomization

[riccardosavorgnan]

No description provided.

[greptile-apps]

Greptile Summary

Fixed reward randomization bounds for reward_bound_collision_max and reward_bound_offroad_max from -2.9 to -0.1, expanding the randomization range from 0.1 to 2.9 for meaningful reward conditioning.

• Collision reward bounds now properly randomize from -3.0 (severe penalty) to -0.1 (minimal penalty)
• Offroad reward bounds now properly randomize from -3.0 (severe penalty) to -0.1 (minimal penalty)
• Training timesteps increased from 2B to 5B (not mentioned in PR title)

Confidence Score: 4/5

• This PR is safe to merge with the reward randomization fix being correct and well-justified
• The reward bounds correction is sound - expanding from a negligible 0.1 range to a meaningful 2.9 range enables proper reward conditioning. Minor point deducted for undocumented total_timesteps increase, though this doesn't introduce bugs
• No files require special attention

Important Files Changed

Filename	Overview
pufferlib/config/ocean/drive.ini	Corrected reward randomization bounds for collision/offroad from narrow range (0.1) to meaningful range (2.9); also increased total_timesteps from 2B to 5B

_{Last reviewed commit: 084327d}

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}