Skip to main content
openrlbenchmark
Projects
sbx
Reports
CrossQ- SBX Perf Report

CrossQ- SBX Perf Report

Antonin RAFFIN
Created on March 29|Last edited on October 7
Note: using the recommended gamma=0.98 for PyBullet envs
Only 3 random seeds so far but the variance is quite low.
The only env where it is inconsistent is Swimmer-v4.
It also underperforms when compared to TQC on BipedalWalkerHardcore-v3.

PyBullet Envs

MuJoCo Envs

HalfCheetah-v4


200k400k600k800k1Mglobal_step02000400060008000100001200014000
algo: sac, env: HalfCheetah-v3 SB3
algo: tqc, env: HalfCheetah-v3 SB3
algo: crossq, env: HalfCheetah-v4, hyperparams.gamma: 0.99 SBX
algo: crossq, env: HalfCheetah-v4, hyperparams.gamma: 0.98 SBX
SBX
6
SB3
20


Ant-v4


SBX
6
SB3
20


Hopper-v4


SBX
6
SB3
20


Walker2d-v4


SBX
6
SB3
20


Created with ❤️ on Weights & Biases.

https://wandb.ai/openrlbenchmark/sbx/reports/CrossQ-SBX-Perf-Report--Vmlldzo3MzQxOTAw
Made with Weights & Biases. Sign up or log in to create reports like this one.