JAT Open-Source Reinforcement Learning Model - Empowered by Multi-Modality and Multi-Task, Performs Exceptionally in Multiple Game Environments

Jat

Developed by jat-project

JAT is a multimodal, multi-task reinforcement learning model that excels in various environments such as Atari games, BabyAI, MetaWorld, and MuJoCo.

Multimodal Fusion

Transformers

Other#Multi-task Reinforcement Learning #Atari Game Control #BabyAI Task Solving

Downloads 71

Release Time : 1/16/2024

Model Overview

JAT is a general-purpose reinforcement learning model capable of handling diverse tasks and environments, including gaming, robot control, and navigation.

Model Features

Multi-task Learning

Capable of performing excellently on multiple different reinforcement learning tasks and environments simultaneously

High Versatility

Applicable to various reinforcement learning scenarios, from gaming to robot control

High Performance

Achieves or approaches expert-level performance in multiple benchmark tests

Model Capabilities

Atari Game Control

BabyAI Task Solving

MetaWorld Robot Manipulation

MuJoCo Physics Simulation Control

Use Cases

Game AI

Atari Game Player

Automatically plays various classic Atari games

IQM human-normalized total reward reaches 0.38

Robot Control

MuJoCo Ant Control

Controls the ant robot in the MuJoCo simulation environment

IQM expert-normalized total reward reaches 0.85

Navigation Tasks

BabyAI Task Solving

Solves various navigation and object manipulation tasks in the BabyAI environment

IQM expert-normalized total reward reaches 0.99

🚀 jat-project/jat

This project is focused on reinforcement learning, with applications in various environments such as Atari, BabyAI, MetaWorld, and MuJoCo. It provides performance metrics on multiple datasets, demonstrating its effectiveness in different reinforcement - learning scenarios.

📚 Documentation

Model Information

Property	Details
Model Name	jat-project/jat
Tags	reinforcement - learning, atari, babyai, metaworld, mujoco - ant, mujoco
Datasets	jat - project/jat - dataset
Pipeline Tag	reinforcement - learning

Results on Different Datasets

Atari 57

Task: Reinforcement Learning
Metrics:
- IQM expert normalized total reward: 0.14 [0.14, 0.15]
- IQM human normalized total reward: 0.38 [0.37, 0.39]

BabyAI

Task: Reinforcement Learning
Metrics:
- IQM expert normalized total reward: 0.99 [0.99, 0.99]

MetaWorld

Task: Reinforcement Learning
Metrics:
- IQM expert normalized total reward: 0.65 [0.64, 0.67]

MuJoCo

Task: Reinforcement Learning
Metrics:
- IQM expert normalized total reward: 0.85 [0.83, 0.86]

Alien (Atari - Alien)

Task: Reinforcement Learning
Metrics:
- Total reward: 1518.70 +/- 568.14
- Expert normalized total reward: 0.08 +/- 0.03
- Human normalized total reward: 0.19 +/- 0.08

Amidar (Atari - Amidar)

Task: Reinforcement Learning
Metrics:
- Total reward: 89.17 +/- 78.73
- Expert normalized total reward: 0.04 +/- 0.04
- Human normalized total reward: 0.05 +/- 0.05

Assault (Atari - Assault)

Task: Reinforcement Learning
Metrics:
- Total reward: 1676.91 +/- 780.73
- Expert normalized total reward: 0.09 +/- 0.05
- Human normalized total reward: 2.80 +/- 1.50

Asterix (Atari - Asterix)

Task: Reinforcement Learning
Metrics:
- Total reward: 844.50 +/- 546.85
- Expert normalized total reward: 0.18 +/- 0.16
- Human normalized total reward: 0.08 +/- 0.07

Asteroids (Atari - Asteroids)

Task: Reinforcement Learning
Metrics:
- Total reward: 1357.90 +/- 453.01
- Expert normalized total reward: 0.00 +/- 0.00
- Human normalized total reward: 0.01 +/- 0.01

Atlantis (Atari - Atlantis)

Task: Reinforcement Learning
Metrics:
- Total reward: 51843.00 +/- 123857.07
- Expert normalized total reward: 0.13 +/- 0.40
- Human normalized total reward: 2.41 +/- 7.66

Bank Heist (Atari - Bankheist)

Task: Reinforcement Learning
Metrics:
- Total reward: 977.80 +/- 156.49
- Expert normalized total reward: 0.74 +/- 0.12
- Human normalized total reward: 1.30 +/- 0.21

Battle Zone (Atari - Battlezone)

Task: Reinforcement Learning
Metrics:
- Total reward: 16780.00 +/- 6926.15
- Expert normalized total reward: 0.06 +/- 0.02
- Human normalized total reward: 0.45 +/- 0.19

Beam Rider (Atari - Beamrider)

Task: Reinforcement Learning
Metrics:
- Total reward: 768.36 +/- 364.06
- Expert normalized total reward: 0.01 +/- 0.01
- Human normalized total reward: 0.02 +/- 0.02

Berzerk (Atari - Berzerk)

Task: Reinforcement Learning
Metrics:
- Total reward: 616.20 +/- 296.08
- Expert normalized total reward: 0.01 +/- 0.01
- Human normalized total reward: 0.20 +/- 0.12

Bowling (Atari - Bowling)

Task: Reinforcement Learning
Metrics:
- Total reward: 22.32 +/- 5.18
- Expert normalized total reward: 1.00 +/- 0.00
- Human normalized total reward: - 0.01 +/- 0.04

Boxing (Atari - Boxing)

Task: Reinforcement Learning
Metrics:
- Total reward: 92.31 +/- 18.24
- Expert normalized total reward: 0.94 +/- 0.19
- Human normalized total reward: 7.68 +/- 1.52

Breakout (Atari - Breakout)

Task: Reinforcement Learning
Metrics:
- Total reward: 7.93 +/- 5.66
- Expert normalized total reward: 0.01 +/- 0.01
- Human normalized total reward: 0.22 +/- 0.20

Centipede (Atari - Centipede)

Task: Reinforcement Learning
Metrics:
- Total reward: 5888.27 +/- 2594.62
- Expert normalized total reward: 0.40 +/- 0.27
- Human normalized total reward: 0.38 +/- 0.26

Chopper Command (Atari - Choppercommand)

Task: Reinforcement Learning
Metrics:
- Total reward: 2371.00 +/- 1195.43
- Expert normalized total reward: 0.02 +/- 0.01
- Human normalized total reward: 0.24 +/- 0.18

Crazy Climber (Atari - Crazyclimber)

Task: Reinforcement Learning
Metrics:
- Total reward: 97145.00 +/- 30388.04
- Expert normalized total reward: 0.51 +/- 0.18
- Human normalized total reward: 3.45 +/- 1.21

Defender (Atari - Defender)

Task: Reinforcement Learning
Metrics:
- Total reward: 39317.50 +/- 16246.15
- Expert normalized total reward: 0.10 +/- 0.05
- Human normalized total reward: 2.30 +/- 1.03

Demon Attack (Atari - Demonattack)

Task: Reinforcement Learning
Metrics:
- Total reward: 795.10 +/- 982.55
- Expert normalized total reward: 0.01 +/- 0.01
- Human normalized total reward: 0.35 +/- 0.54

Double Dunk (Atari - Doubledunk)

Task: Reinforcement Learning
Metrics:
- Total reward: 13.40 +/- 11.07
- Expert normalized total reward: 0.81 +/- 0.28
- Human normalized total reward: 0.91 +/- 0.32

Enduro (Atari - Enduro)

Task: Reinforcement Learning
Metrics:
- Total reward: 103.11 +/- 28.05
- Expert normalized total reward: 0.04 +/- 0.01
- Human normalized total reward: 0.12 +/- 0.03

Fishing Derby (Atari - Fishingderby)

Task: Reinforcement Learning
Metrics:
- Total reward: - 31.67 +/- 22.54
- Expert normalized total reward: 0.61 +/- 0.23
- Human normalized total reward: 0.46 +/- 0.17

Freeway (Atari - Freeway)

Task: Reinforcement Learning
Metrics:
- Total reward: 27.57 +/- 1.87
- Expert normalized total reward: 0.81 +/- 0.06
- Human normalized total reward: 0.93 +/- 0.06

Frostbite (Atari - Frostbite)

Task: Reinforcement Learning
Metrics:
- Total reward: 2875.60 +/- 1679.84
- Expert normalized total reward: 0.21 +/- 0.13
- Human normalized total reward: 0.66 +/- 0.39

Gopher (Atari - Gopher)

Task: Reinforcement Learning
Metrics:
- Total reward: 5508.80 +/- 2802.03
- Expert normalized total reward: 0.06 +/- 0.03
- Human normalized total reward: 2.44 +/- 1.30

Gravitar (Atari - Gravitar)

Task: Reinforcement Learning
Metrics:
- Total reward: 1330.50 +/- 918.23
- Expert normalized total reward: 0.30 +/- 0.24
- Human normalized total reward: 0.36 +/- 0.29

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご