Our enhancements to PBT allow us to jointly adapt architectures and hyperparameters on the fly and discover entirely new behaviors on the Brax environment suite, which we display here!
HalfCheetah (PPO)
HalfCheetah (BG-PBT)
Ant (PPO)
Ant (BG-PBT)
Hopper (PPO)
Hopper (BG-PBT)
Humanoid (PPO)
Humanoid (BG-PBT)
UR5e (PPO)
UR5e (BG-PBT)
Reacher (PPO)
Reacher (BG-PBT)