pSim 0.2.2

A pythonic Very Small Size Soccer (VSSS) simulation environment for reinforcement learning research.

pSim Documentation

pSim: Very Small Size Soccer Simulation Environment

Comprehensive simulation platform for autonomous robotic soccer development

---

Welcome to pSim

pSim is a comprehensive, pythonic simulation environment specifically designed for the Very Small Size Soccer (VSSS) competition. It provides researchers, students, and developers with powerful tools to develop, test, and validate autonomous robotic soccer strategies and algorithms.

Whether you're working on traditional control algorithms, reinforcement learning, or multi-agent coordination, pSim offers the flexibility and performance you need.

---

Quick Start

Installation

pSim uses uv for fast, reliable package management. See the installation guide for complete setup instructions including uv installation.

# Quick install with uv
uv pip install pSim

# Or with optional dependencies
uv pip install pSim[gym]   # For Gymnasium support
uv pip install pSim[zoo]   # For PettingZoo support
uv pip install pSim[all]   # For all optional dependencies

First Simulation

from pSim import SimpleVSSSEnv
import numpy as np

# Create your first simulation with custom robot counts
env = SimpleVSSSEnv(
    render_mode="human",
    scenario="formation",
    num_agent_robots=2,      # Number of controllable robots
    num_adversary_robots=3   # Number of opponent robots
)
obs, info = env.reset()

# Run simulation steps with random actions
for step in range(1000):
    # Generate random actions for all agent robots [v, w] (velocity, angular velocity)
    action = np.random.uniform(-1, 1, (env.num_agent_robots, 2))
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()

env.close()

---

Environment Types

pSim provides three main environment types, each designed for different use cases:

SimpleEnv - Traditional Control

Perfect for traditional control algorithms, manual testing, and educational purposes:

from pSim import SimpleVSSSEnv
import numpy as np

env = SimpleVSSSEnv(
    render_mode="human",
    scenario="formation",
    num_agent_robots=3,
    num_adversary_robots=3
)

obs, info = env.reset()
for step in range(1000):
    # Generate random actions [v, w] for each controllable robot
    action = np.random.uniform(-1, 1, (env.num_agent_robots, 2))
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()

Manual Control with Joystick

from pSim import SimpleVSSSEnv, HMI

hmi = HMI()

env = SimpleVSSSEnv(
    render_mode="human",
    scenario="formation",
    num_agent_robots=3,
    num_adversary_robots=3
)

obs, info = env.reset()
while hmi.active:
    actions, reset, exit_requested = hmi()

    if exit_requested:
        break
    if reset:
        obs, info = env.reset()
        continue

    obs, reward, terminated, truncated, info = env.step(actions)

    if terminated or truncated:
        obs, info = env.reset()

hmi.quit()
env.close()

VSSSGymEnv - Reinforcement Learning

Full Gymnasium compatibility for single-agent reinforcement learning:

from pSim import VSSSGymEnv
from gymnasium.wrappers import FlattenObservation

env = FlattenObservation(VSSSGymEnv(
    render_mode="human",
    scenario="formation",
    num_agent_robots=3,
    num_adversary_robots=3
))

obs, info = env.reset()
for step in range(1000):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()

VSSSPettingZooEnv - Multi-Agent Learning

PettingZoo interface for multi-agent reinforcement learning:

from pSim import VSSSPettingZooEnv

env = VSSSPettingZooEnv(
    render_mode="human",
    scenario="formation",
    num_agent_robots=3,
    num_adversary_robots=3
)

obs, info = env.reset()
for step in range(1000):
    actions = {agent: env.action_space(agent).sample() for agent in env.agents}
    obs, rewards, terminations, truncations, infos = env.step(actions)
    if any(terminations.values()) or any(truncations.values()):
        obs, info = env.reset()

---

Key Features

Flexible Robot Control

Easy-to-use JSON-based configuration for scenarios, robot behaviors, and simulation parameters. Mix controllable agents with automatic behaviors:

{
  "scenarios": {
    "formation": {
      "agent_robots": {
        "movement_types": ["action", "ou", "no_move"]
      },
      "adversary_robots": {
        "movement_types": ["ou", "ou", "ou"]
      }
    }
  }
}

• "action": Controllable by your agent/algorithm
• "ou": Ornstein-Uhlenbeck automatic movement
• "no_move": Stationary robots

Realistic Physics

Box2D-powered physics simulation with customizable parameters for accurate robot and ball dynamics.

Human-Machine Interface

• Keyboard Controls: Full keyboard input with intuitive mappings
• Joystick Support: Universal controller compatibility
• Robot Switching: Dynamic selection between controllable robots
• Team Management: Control different teams independently
• Ball Control Mode: Direct ball manipulation

---

Use Cases

Traditional Control

• PID controllers, MPC, and other traditional methods
• Path planning and trajectory optimization
• Formation control and cooperative behaviors

Reinforcement Learning

• Compatible with Stable Baselines 3, Ray RLlib, and other RL libraries
• Customizable reward functions and observation spaces
• Curriculum learning through scenario configuration

Research & Education

• Multi-agent coordination studies
• Emergent behavior analysis
• Algorithm benchmarking and comparison

Competition Preparation

• Strategy development and testing
• Opponent modeling and adaptation
• Performance analysis and optimization

---

Documentation Structure

Getting Started

• Installation: Complete setup instructions
• Usage Guide: Comprehensive usage examples and tutorials
• Configuration: Detailed configuration options

Examples

• Configuration Demo: Setting up robot behaviors
• HMI Control: Manual control examples
• Simple Environment: Traditional control examples
• Gym Environment: Reinforcement learning examples
• PettingZoo Environment: Multi-agent examples

API Reference

• API Documentation: Complete API reference

---

Getting Help

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Documentation: Full Documentation

---

Contributing

We welcome contributions! Please see our Contributing Guide for details.

---

Ready to start developing robotic soccer strategies?

Get Started View Examples GitHub Repository