config_examples

A repository that contains example json files that can be used for different applications of the LeRobot code base.

Current available configs:

- RL :
- env_config.json: Environment config for a real robot setup using a gamepad for teleoperation and an SO10* arm as the main agent. Using `gym_manipulator.py`, one can use this config to teleoperate the robot and record a dataset for reinforcement learning.
- train_config.json: Training config for the HIL-SERL implementation in LeRobot on the real robot using the similar environment configuration to the `env_config.json` in the same directory.
- gym_hil:
- env_config.json: Environment config for simulation using the `gym_hil` environment.
- train_config.json: Training config for the HIL-SERL implementation in LeRobot for simulated environments.
- Sim IL:
- env_config.json: Environment config for simulation using the `gym_hil` environment. You can use this configuration to collect a dataset in simulation that can be used for imitation learning or reinforcement learning.
- eval_config.json: Evaluation config for models trained on datasets collected from `gym_hil` environment.
- Reward Classifier:
- config.json: Main configuration for training a reward classifier.

本仓库包含可适配LeRobot代码库各类应用场景的示例JSON文件。

当前可用配置如下：
- RL：
- env_config.json：适用于以游戏手柄实现遥操作、以SO10*机械臂作为主控智能体的实体机器人搭建环境配置文件。结合`gym_manipulator.py`，可通过该配置完成机器人遥操作并录制用于强化学习的数据集。
- train_config.json：针对LeRobot中HIL-SERL实现的实体机器人训练配置文件，其环境配置与同目录下的`env_config.json`保持一致。
- gym_hil：
- env_config.json：基于`gym_hil`环境的仿真场景配置文件。
- train_config.json：针对LeRobot中HIL-SERL实现的仿真环境训练配置文件。
- Sim IL：
- env_config.json：基于`gym_hil`环境的仿真场景配置文件。可通过该配置在仿真环境中采集数据集，用于模仿学习或强化学习任务。
- eval_config.json：针对基于`gym_hil`环境采集的数据集训练得到的模型的评估配置文件。
- 奖励分类器（Reward Classifier）：
- config.json：用于训练奖励分类器的核心配置文件。