Examples

Environments

• Getting Started
• Satellite Configuration
• Multi-Agent Environments
• Fault Environment Example
• Configure Satellite Pareameters
• Broadcast Communication
• Cloud Environment

Earth Observation

• Cloud Environment
• Cloud Environment with Re-imaging
• Agile Earth-Observing Satellite Environment

RSO Inspection

• RSO Inspection

Training

• Training with RLlib PPO
• Time-Discounted GAE
• Asynchronous Multiagent Decision Making
• Training with Curriculum Learning
• Shielded training with action masking and action replacement

Benchmarks

BSK-RL includes benchmark environments that can be used for training and evaluating RL algorithms. These can be found in the benchmarks directory and trained using PPO with

python bsk_rl/benchmarks/benchmark.py -o results_dir -e nadir_science:nadir_science

To see a full list of options for the training script, run

python bsk_rl/benchmarks/benchmark.py -h

Environments are specified in the format [file_name]:[env_name], where file_name is the name of a Python file in the benchmarks directory and env_name is the name of an environment defined in that file. The environment includes both simulation and training settings. The following environments are currently available:

• nadir_science:nadir_science: A simple science environment with resource constraints.

• aeos:aeos_single: A single-satellite agile Earth observation environment.

• aeos:aeos_constellation: A multi-satellite agile Earth observation environment.

• rso_inspection:rso_inspection: An RSO inspection environment with imaging constraints and safety constraints.