whereismyai
Simulusion
Simulusion is a modular PyTorch library that brings major world-model families together under a single consistent API.
Simulusion is a modular PyTorch library designed for learning, training, and deploying world models across various environments. It brings major world-model families together under a single consistent API, allowing users to train algorithms like Dreamer, PlaNet, JEPA, IRIS, DIAMOND, DiT, and Genie. Users can use the top-level API or drop down to encoders, decoders, and latent-dynamics backbones to compose custom architectures. The library includes built-in environment adapters for Gym, Gymnasium, DeepMind Control, MuJoCo, Brax, Atari, and Unity ML-Agents, alongside ONNX, TorchScript, and TensorRT export. It is built for researchers and developers working on model-based reinforcement learning, self-supervised visual representations, and generative interactive environments.
Simulusion at a glance
Key features
- Unified interfaces across world-model algorithms with a single protocol and trainer
- Single episode format for video, reinforcement learning, and robot data with a LeRobot reader
- Modular encoders, decoders, dynamics models, and backbones
- Training and inference utilities for model-based reinforcement learning
- Environment integrations for Gym, Gymnasium, Unity ML-Agents, MuJoCo, and Brax
- Export support for ONNX, TorchScript, and TensorRT
- Shared YAML run configuration for all algorithm families
Use it for
- Train and compare multiple world-model algorithms with a single API
- Export trained world models to ONNX, TorchScript, or TensorRT
- Train world models from recorded episodes and robot data
- Run model-based reinforcement learning across various simulator environments
Alternatives to Simulusion
- Stable Baselines3 — Provides reliable, standard PyTorch implementations for reinforcement learning.
- NeMo Molt — It scales fully-async agentic RL to frontier MoE scale with a small PyTorch stack.
- RLinf — Scalable open-source framework supporting real-world robots and simulators.
- OpenVLHarness — Lifts six multimodal model backbones by 9.7 to 12.6 points across 22 benchmarks.
- Karotte — It runs tasks in isolated virtual machines for safety and consistency.
- Seahaven — Provides isolated, reproducible, and stateful synthetic environments for agents.
- Finetuning With Sampling — Official implementation of the Finetuning with Sampling research paper from Harvard.
- Kiln AI — Provides stateful, resettable, and parallel environments for agent evals.
- MiroFlow — Achieves top rankings on multiple agentic benchmarks with modular architecture.
- Dreamscale Labs — Run frontier robot models in the cloud without local GPUs.
- Pythos — Provides Socratic guidance and verified derivations for math and physics.
- Vegaml — Runs locally on your hardware with calibrated probabilities and conformal answer sets.
All Simulusion alternatives