Demos & Examples

JoltGym ships with several example scripts in the examples/ directory covering basic usage, RL training, and video recording.

Available Examples

Script	Description
benchmark.py	Unified benchmark: train + record + throughput for any env
demo_halfcheetah.py	Random-action HalfCheetah demo
demo_vectorized.py	WorldPool throughput benchmark
train_ppo.py	Train HalfCheetah with PPO
train_humanoid.py	Train Humanoid with PPO
train_multiagent.py	Train CheetahRace with PPO
record_video.py	Record HalfCheetah video
record_humanoid.py	Record Humanoid video
record_race.py	Record multi-agent race video

Basic Demo

examples/demo_halfcheetah.py

Runs 1000 steps of HalfCheetah with random actions, printing observations and rewards:

python examples/demo_halfcheetah.py

import joltgym

env = joltgym.make("JoltGym/HalfCheetah-v0")
obs, info = env.reset(seed=42)

total_reward = 0
for step in range(1000):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    total_reward += reward

    if step % 100 == 0:
        print(f"  Step {step:4d} | reward={reward:7.3f} | "
              f"x_pos={info['x_position']:7.3f} | "
              f"x_vel={info['x_velocity']:7.3f}")

    if terminated or truncated:
        obs, info = env.reset()
        total_reward = 0

env.close()

Vectorized Benchmark

examples/demo_vectorized.py

Benchmarks the C++ WorldPool with 64 parallel environments over 10,000 steps:

python examples/demo_vectorized.py

import numpy as np
from joltgym.vector.jolt_vector_env import JoltVectorEnv

num_envs = 64
envs = JoltVectorEnv(num_envs, model_path="python/joltgym/assets/half_cheetah.xml")
obs, info = envs.reset(seed=42)

import time
start = time.time()
for step in range(10000):
    actions = np.random.uniform(-1, 1, (num_envs, 6)).astype(np.float32)
    obs, rewards, terms, truncs, infos = envs.step(actions)
elapsed = time.time() - start

print(f"Throughput: {num_envs * 10000 / elapsed:,.0f} env-steps/second")

Prerequisites

All demos require JoltGym to be installed:

pip install -e .

Training demos additionally require:

pip install stable-baselines3 tensorboard

Recording demos additionally require:

pip install matplotlib