NavigationController

Spawn Test Environment

from simulations.maze_env import MazeEnv
from simulations.point import PointEnv
from simulations.maze_task import CustomGoalReward4Rooms
env = MazeEnv(PointEnv, CustomRewardGoal4Rooms)

Experiment Checklist

• Check and update def step(**kwargs) and def _set_action_space(**kwargs) in simulations/point.py
• Check and update def _get_obs(**kwargs) and def _set_observation_space(**kwargs) in simulations/maze_env.py and simulations/point.py
• Update Reward and Info Computation Pipeline
• Check and update class Actor and class Critic in utils/rtd3_utils.py
• Check and update class Autoencoder, class VisualCortexV4, class EncoderBody and class FeatureExtractionBackbone in bg/models.py
• Check and update def update_policy(**kwargs) and def _sample_action(**kwargs) in utils/rtd3_utils.py
• Check and update def predict(**kwargs) in utils/rtd3_utils.py
• Ensure squash_output is False in class Actor if using inverted gradients method
• Check and update seed, batch_size, learning_starts, imitation_steps and debug in params from constant.py
• Check and update other hyperparameters for tuning
• Set LOGDIR, ENV_TYPE, TIMESTEPS, MAX_EPISODE_SIZE, HISTORY_STEPS and TASK_VERSION in train.sh
• Check and update net_arch and n_critics in learning/explore.py
• Run sh train.sh in terminal for debugging. Run nohup sh train.sh >> assets/out/models/train.log & in terminal for GPU execution.
• Commit and Push to Github. Update local experiments_log

Installation on docker

• Run docker build -t neuroengineering-tools ./docker/cpu/ in the root folder of the repository.
• Run docker run -dit --name sample neuroengineering-tools to spawn a new container.
• Run desired command with the tools in the container.
• Run docker stop neuroengineering-tools to stop running container.