Add benchmark environments

[Mark2000]

Description

Closes #339

Adds benchmark environments. Run with:

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

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)

How should this pull request be reviewed?

• By commit
• All changes at once

Future Work

Test benchmarks automatically: #340

Checklist

• I have performed a self-review of my code
• I have commented my code in hard-to-understand areas
• I have made corresponding changes to the documentation and release notes
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works

[Copilot]

Pull request overview

Adds a lightweight benchmark harness (env templates + a Ray RLlib PPO training entrypoint) to support repeatable training runs against specific benchmark environments, and fixes a Basilisk support-data path fallback bug.

Changes:

• Fix Basilisk dataFetcher ImportError fallback path variable naming in WorldModel setup.
• Add benchmarks/train.py RLlib PPO training script with checkpointing and dynamic benchmark loading.
• Add a first benchmark environment (nadir_science) plus a small Benchmark dataclass template.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 6 comments.

File	Description
src/bsk_rl/sim/world.py	Fixes fallback Basilisk path variable used when dataFetcher API is unavailable.
benchmarks/train.py	New RLlib PPO training/continue script for benchmark runs (checkpoint mgmt, dynamic env import).
benchmarks/nadir_science.py	Defines a benchmark environment configuration (satellite model + env/training args).
benchmarks/env_template.py	Introduces a simple Benchmark dataclass container for env/training configuration.

Comments suppressed due to low confidence (2)

benchmarks/train.py:359

• The script initializes Ray and a PPO algorithm but never calls ppo.stop() / ray.shutdown() before exiting. This can leave worker processes running (especially in interactive or repeated benchmark runs) and can hold onto temp dirs/object store resources. Consider adding a try/finally around training to ensure cleanup.
  benchmarks/nadir_science.py:133
• These print(...) statements will execute at import time, which is noisy when using the dynamic import in benchmarks/train.py and makes it harder to use these benchmarks as a library. Consider removing them or switching to logging guarded by if __name__ == "__main__" / a verbosity flag.

training_args = dict(
    lr=3e-5,
    gamma=0.999,
    train_batch_size=1000,
    num_sgd_iter=10,
    use_kl_loss=False,
    clip_param=0.1,
    grad_clip=0.5,
)

nadir_science_benchmark = Benchmark(
    env_args=env_args,
    policies=policies,
    policy_mapping_fn=policy_mapping_fn,
    module_specs=module_specs,
    training_args=training_args,
)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 13 out of 14 changed files in this pull request and generated 4 comments.