SuperDex Gym

SuperDex Gym is a reinforcement-learning (RL) framework built on the SuperDex Physics engine. It provides an environment Python API that presents a simulated task through Gymnasium, a standard Python interface for reinforcement learning. You can use these environments to train AI agent controllers in high-fidelity physics simulations.

What to Install

SuperDex Lab delivers installable environments and source-only app scripts in different ways. An app script is an executable Python program for training or benchmarking.

1. To use an environment, install the superdex-lab package. The package provides Python environment classes for common robot agents. These classes implement the Gymnasium interface and communicate with the SuperDex Physics C++ simulation backend.
2. To train with Ray/RLlib or run benchmarks, use the source checkout. A source checkout is a local copy of the Project SuperDex repository; superdex_lab is a subproject within it. The app scripts are not part of the installable package. The Ray/RLlib integration in apps/rllib/ provides quick access to reinforcement learning algorithms and efficient distributed training. The benchmark entry point is apps/envs/benchmark.py.

A core source sync does not provide all dependencies for the source-only app scripts. Install tqdm for the app scripts, plus the training stack for RLlib work. See dependencies for apps/.

SuperDex Gym includes classic-control and locomotion benchmark environments. All of them use the SuperDex Physics engine for high-fidelity simulation.

Key Features

• High-fidelity Physics: Powered by the SuperDex Physics engine for accurate simulations
• Gymnasium Compatible: Standard reinforcement-learning interface for easy integration with existing workflows
• Ray/RLlib Integration: Distributed training with Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC)
• Environment Suite: Classic-control and locomotion benchmark environments
• Flexible Configuration: Extensive configuration options for environments and training
• Visualization: Built-in Polyscope rendering and video generation
• Cross-platform Support: Runs on Linux, macOS and Windows

Available Environments

Each environment has two identifiers:

• The command-line interface (CLI) name is the snake_case value passed to command-line tools, such as cart_pole.
• The Gymnasium ID is the registered identifier passed to gym.make(), such as superdex_gym/CartPole-v0.

Environment	CLI name	Gymnasium ID	Description
Ant	ant	superdex_gym/Ant-v0	Quadrupedal locomotion task benchmark
CartPole	cart_pole	superdex_gym/CartPole-v0	Classic pole balancing task benchmark
HalfCheetah	half_cheetah	superdex_gym/HalfCheetah-v0	Planar running task benchmark

Base environments come from *_env.py modules. Sibling JSON files define configuration variants and recipes. This catalog reflects a source checkout; follow setup for the install route.

To print the exact environments available in your install, run this command from the superdex_lab project root:

uv run python apps/envs/run_sample.py --help

Python adds the script's own directory to sys.path, so the script's sibling imports resolve regardless of the directory from which you invoke it.

A config variant is an alternative configuration of an existing environment class. Each registered variant has its own Gymnasium ID.

Variant CLI name	Gymnasium ID
ant_full_observation	superdex_gym/AntFullObservation-v0
ant_no_contact	superdex_gym/AntNoContact-v0
ant_rotation_vector	superdex_gym/AntRotationVector-v0
cart_pole_actuate_on_pole	superdex_gym/CartPoleActuateOnPole-v0
half_cheetah_full_observation	superdex_gym/HalfCheetahFullObservation-v0

See Environment file naming for how variants are declared. Variants with a test segment in their names are smoke-tested but never registered, so they never appear in a CLI listing.

Environment prerequisites

The included cart_pole, ant and half_cheetah environments load their scenes and meshes from assets/benchmarks/ from the Project Superdex source code on Github.

All environments require superdex-robotics, which should be installed as a dependency of superdex-lab. See Verifying the Install.