体现

快速强化学习研究。

概述

Embodied的目标是赋予研究人员快速大规模实现新代理的能力。Embodied通过为环境和代理指定接口来实现这一点，允许用户混合匹配代理、环境和评估协议。Embodied提供了一些常见的构建块，鼓励用户在需要更多控制时进行分支。唯一的依赖项是Numpy，代理可以在任何框架中实现。

包

embodied/
  core/    # Config, logging, checkpointing, simulation, wrappers
  run/     # Evaluation protocols that combine agents and environments
  envs/    # Environment suites such as Gym, Atari, DMC, Crafter
  agents/  # Agent implementations

代理API

class Agent:
  __init__(obs_space, act_space, config)
  policy(obs, carry, mode='train') -> act, carry
  train(data, carry) -> metrics, carry
  report(data, carry) -> metrics, carry
  init_policy(batch_size) -> carry
  init_train(batch_size) -> carry
  init_report(batch_size) -> carry
  dataset(generator) -> generator

环境API

class Env:
  __len__() -> int
  @obs_space -> dict of spaces
  @act_space -> dict of spaces
  step(act) -> obs dict
  render() -> array
  close()