Continuous Control
Continuous Control是指在游戏等环境中,通过一系列平滑、持续的调整或动作来实现精准控制的能力。其目标是在需要精确度、时机和动作幅度的场景中,优化决策过程和执行效果。Continuous Control在赛车游戏、角色模拟和飞行模拟器等应用中具有重要价值,能够提升系统的响应性和灵活性,增强用户体验和系统性能。
2D Walker
Acrobot
Acrobot (limited sensors)
Acrobot (noisy observations)
acrobot.swingup
SMuZero
Acrobot (system identifications)
Ant
Ant + Gathering
Ant + Maze
ball_in_cup.catch
Ball in cup, catch (DMControl100k)
Ball in cup, catch (DMControl500k)
Cart-Pole Balancing
TRPO
Cart-Pole Balancing (limited sensors)
Cart-Pole Balancing (noisy observations)
Cart-Pole Balancing (system identifications)
Cart Pole (OpenAI Gym)
MAC
cartpole.balance
cartpole.balance_sparse
cartpole.swingup
Cartpole, swingup (DMControl100k)
Cartpole, swingup (DMControl500k)
cartpole.swingup_sparse
cheetah.run
Cheetah, run (DMControl100k)
Cheetah, run (DMControl500k)
DeepMind Cheetah Run (Images)
DrQ
DeepMind Cup Catch (Images)
DrQ
DeepMind Walker Walk (Images)
DrQ
Double Inverted Pendulum
finger.spin
Finger, spin (DMControl100k)
Finger, spin (DMControl500k)
CURL
finger.turn_easy
finger.turn_hard
fish.swim
Full Humanoid
Half-Cheetah
Hopper
hopper.hop
hopper.stand
humanoid.run
Inverted Pendulum
TRPO
Inverted Pendulum (limited sensors)
Inverted Pendulum (noisy observations)
Inverted Pendulum (system identifications)
Lunar Lander (OpenAI Gym)
SAC
manipulator.insert_ball
manipulator.insert_peg
Mountain Car
Mountain Car (limited sensors)
Mountain Car (noisy observations)
Mountain Car (system identifications)
pendulum.swingup
PyBullet Ant
TD3 gSDE
PyBullet HalfCheetah
SAC
PyBullet Hopper
PyBullet Walker2D
quadruped.run
quadruped.walk
reacher.easy
Reacher, easy (DMControl100k)
Reacher, easy (DMControl500k)
reacher.hard
Simple Humanoid
Swimmer
Swimmer + Gathering
Swimmer + Maze
walker.run
walker.stand
walker.walk
Walker, walk (DMControl100k)
Walker, walk (DMControl500k)
CURL