2024 • Lightweight Social Navigation Environment • Optimized for Rapid Iteration
I developed this lightweight version of SocNavGym optimized for rapid iteration and faster simulation. Designed for efficient testing of predictive models and reinforcement learning approaches, it maintains core social interaction dynamics while dramatically reducing simulation time.
Watch LiteSocNavGym in action: faster simulations with core social navigation features
While SocNavGym provides a comprehensive social navigation environment, I observed that it could be computationally expensive for rapid iteration during algorithm development. When testing predictive models and reinforcement learning approaches, I needed faster feedback loops.
I developed LiteSocNavGym as a lite testing environment to enable rapid iteration during algorithm development. It allows for quick testing of predictive models and reinforcement learning approaches before applying methods to the full SocNavGym environment.
I created LiteSocNavGym by simplifying complex features while preserving core social navigation dynamics. The environment achieves 3-5x faster simulation speeds by optimizing rendering, physics calculations, and state management (see Key Innovation section for details).
Creating a lightweight version of SocNavGym required careful balance:
I addressed these challenges by:
LiteSocNavGym achieves 3-5x faster simulation speeds by:
LiteSocNavGym achieves 3-5x faster simulation speeds compared to SocNavGym while maintaining essential social navigation dynamics (see What Was Preserved section). This performance gain makes it ideal for rapid algorithm development, quick iteration during model training, efficient benchmarking, and prototyping.
Demo Video: LiteSocNavGym Environment
Watch the environment in action: faster simulations with core social navigation features preserved (see What Was Preserved section).
To achieve faster simulation speeds, LiteSocNavGym:
Despite optimizations, LiteSocNavGym maintains:
Use LiteSocNavGym when:
Use SocNavGym when:
import gymnasium as gym
from litesocnavgym import LiteSocNavGym
# Create environment
env = LiteSocNavGym()
# Reset environment
obs, info = env.reset()
# Run episode
for step in range(1000):
# Get action from your model
action = your_model.predict(obs)
# Step environment
obs, reward, terminated, truncated, info = env.step(action)
# Render (optional, faster than SocNavGym)
env.render()
if terminated or truncated:
obs, info = env.reset()
env.close()
Code written for SocNavGym can typically run in LiteSocNavGym with minimal changes:
# SocNavGym code
from socnavgym import SocNavGym
env = SocNavGym()
# LiteSocNavGym code (same structure)
from litesocnavgym import LiteSocNavGym
env = LiteSocNavGym()
# Same API, faster execution
LiteSocNavGym enabled rapid iteration during algorithm development, allowing quick testing before applying methods to the full SocNavGym environment.
Key Contributions:
LiteSocNavGym was used for:
| Feature | LiteSocNavGym | SocNavGym |
|---|---|---|
| Simulation Speed | 3-5x Faster | Baseline |
| Visual Rendering | Lite | Full Detail |
| Core Dynamics | Preserved | Full |
| Use Case | Rapid Development | Final Evaluation |
| API Compatibility | Compatible | Full |
LiteSocNavGym:
SocNavGym:
A: LiteSocNavGym achieves 3-5x faster simulation speeds compared to SocNavGym (see Results & Performance section), making it ideal for rapid algorithm development and testing.
A: Visual rendering was made lite, physics calculations were optimized, and non-essential features were removed (see What Was Simplified section). However, essential social navigation dynamics are preserved (see What Was Preserved section).
A: Yes! LiteSocNavGym maintains API compatibility with SocNavGym, so code written for one can typically run in the other with minimal changes.
A: Use LiteSocNavGym for rapid development, testing, and prototyping. Use SocNavGym for final evaluation, comprehensive testing, and when you need full visual rendering.
A: Yes, essential dynamics are preserved, so results should be comparable. However, always validate final results in the full SocNavGym environment.
A: LiteSocNavGym has been used for testing Predictive World Models (2StepAhead, MASPM) and Adaptive Prediction Horizons (see Research Impact section for details).
A: Install the environment, check the Installation & Usage section above, and start with the basic usage example. The API is compatible with SocNavGym, so existing code should work.
Technologies: Python • Gymnasium • Reinforcement Learning • Social Navigation
Related Projects:
Research Applications: