Principle:Facebookresearch Habitat lab Benchmark Configuration
| Knowledge Sources | |
|---|---|
| Domains | Evaluation, Software_Architecture |
| Last Updated | 2026-02-15 02:00 GMT |
Overview
Initialization of the Habitat Benchmark framework that creates an evaluation environment from configuration for standardized agent testing.
Description
Benchmark Configuration creates a `Benchmark` instance from a config path, initializing the Habitat environment with the specified task, dataset, and simulator settings. The Benchmark handles environment lifecycle, episode iteration, and metric collection, providing a standardized evaluation protocol that ensures reproducible results.
Usage
Use when you want to evaluate an agent using the standard Habitat benchmark protocol. This provides the canonical evaluation setup for comparing agents.
Theoretical Basis
The benchmark pattern provides:
- Standardized evaluation: Same environment setup across all agents
- Reproducibility: Config-driven setup ensures identical conditions
- Metric aggregation: Consistent collection and averaging of metrics
# Abstract benchmark setup
benchmark = Benchmark(config_path)
# Creates env internally, manages episode iteration
metrics = benchmark.evaluate(agent, num_episodes=100)