| name | implementing-random-agent |
| description | Develop a baseline random agent to stress-test the environment and establish minimum performance bounds. |
| version | 0.1.0 |
Implementing the Random Baseline Agent
Before any deep neural networks consume GPU cycles, a perfectly random agent must be deployed inside the finalized Gymnasium environment. This serves two critical purposes: stability testing and establishing the absolute zero-point of intelligence.
Step 1 — Environment Instantiation
Load the custom env.py.
- Apply all intended wrappers (TimeLimit, Observation wrappers).
Step 2 — Constructing the Random Loop
Write a script that queries env.action_space.sample() without replacing any logic.
- Run for exactly
1,000,000 continuous steps.
- Goal: Guarantee the environment does not throw memory leaks, indexing errors, or segmentation faults during extended arbitrary rollouts.
Step 3 — Record Minimum Baseline Score
Track the episode_reward across 1,000 episodes completed by the random agent.
- Calculate the Mean and Standard Deviation.
- This is the "Zero Intelligence" score. If a complex RL algorithm scores lower than this, learning has entirely collapsed.
Output Format
Produce a python script (baseline_random.py) and log the metrics in the output document:
### Random Agent Baseline
- **Hardware Stability**: [Passed 1M steps / Failed with error X]
- **Mean Episode Return**: [-14.5 ± 2.1]
- **Average Episode Length**: [140 steps]
