Skip to main content
tlab
Projects
CORL
Reports
[Offline] ReBRAC

[Offline] ReBRAC

Vladislav Kurenkov, Denis Tarasov
Created on July 16|Last edited on July 16
Results are averaged over 4 seeds. For each dataset we plot d4rl normalized score.
Reference scores are from Revisiting the Minimalist Approach to Offline Reinforcement Learning

Locomotion

Halfcheetah

medium-v2

Reference score: 65.6

2004006008001k1.2kStep0102030405060
Run set
12



medium-replay-v2

Reference score: 51.0

Run set
12



medium-expert-v2

Reference score: 101.1

Run set
12



Walker2d

medium-v2

Reference score: 82.5

Run set
12


medium-replay-v2

Reference score: 77.3

Run set
12


medium-expert-v2

Reference score: 111.6

Run set
12


Hopper

medium-v2

Reference score: 102.0

Run set
12


medium-replay-v2

Reference score: 98.1

Run set
12


medium-expert-v2

Reference score: 107.0

Run set
12


Maze2d

umaze-v1

Reference score: -

Run set
4


medium-v1

Reference score: -

Run set
4


large-v1

Reference score: -

Run set
4


AntMaze

umaze-v2

Reference score: 97.8

Run set
12


umaze-diverse-v2

Reference score: 88.3

Run set
16


medium-play-v2

Reference score: 84.0

Run set
20


medium-diverse-v2

Reference score: 76.3

Run set
20


large-play-v2

Reference score: 60.4

Run set
28


large-diverse-v2

Reference score: 54.4

Run set
12


Adroit

Pen

Human-v1

Reference score: 103.5

Run set
4


Cloned-v1

Reference score: 91.8

Run set
24


Expert-v1

Reference score: 154.1

Run set
28


Door

Human-v1

Reference score: 0.0

Run set
24


Cloned-v1

Reference score: 1.1

Run set
28


Expert-v1

Reference score: 104.6

Run set
36



Hammer

Human-v1

Reference score: 0.2

Run set
24


Cloned-v1

Reference score: 6.7

Run set
28


Expert-v1

Reference score: 133.8

Run set
36



Relocate

Human-v1

Reference score: 0.0

Run set
24


Cloned-v1

Reference score: 0.9

Run set
28


Expert-v1

Reference score: 106.6

Run set
36



Created with ❤️ on Weights & Biases.

https://wandb.ai/tlab/CORL/reports/-Offline-ReBRAC--Vmlldzo0ODkzOTQ2
Made with Weights & Biases. Sign up or log in to create reports like this one.