An open-source quadruped robot dog featuring a dual-core ESP32-S3 architecture and closed-loop servo control for precise movement and interaction.

Evo‑RL is an open‑source, LeRobot‑based framework for real‑world reinforcement learning on Seeed Studio SO‑101 and AgileX PiPER/PiPER‑X robots. It provides end‑to‑end scripts for hardware setup, data collection, value‑function training, advantage‑conditioned policy training, and closed‑loop rollout, with seamless integration to Hugging Face datasets and model hubs.

FlyGym is a Python library that provides a fast, physics‑based digital twin of an adult fruit fly. It simulates detailed biomechanics, compound‑eye vision, olfaction, leg adhesion, and offers a hierarchical brain‑spinal‑cord control interface, enabling researchers to develop and test embodied sensorimotor controllers.

ApeRAG is an open‑source, production‑ready Retrieval‑Augmented Generation platform that combines vector, full‑text, graph, summary and vision search with AI agents. It provides Docker‑Compose quick‑start, full Kubernetes/Helm deployment, MCP integration, and optional advanced document parsing (MinerU). Ideal for building enterprise knowledge‑base assistants, multimodal document search, and AI‑driven support bots.

SparseDrive is an end-to-end autonomous driving framework that uses sparse scene representation to unify perception and planning, significantly improving inference speed and reducing collision rates.

A tool that converts human motion capture data from BVH files into joint animations for humanoid robots by adapting proportions and solving inverse kinematics.

PTAN (PyTorch AgentNet) is a small open‑source library that supplies reusable RL building blocks—experience buffers, training loops, and Ignite integration—for PyTorch‑based agents working with OpenAI‑Gym environments.

Neural Robot Dynamics (NeRD) is a learned dynamics model for articulated rigid bodies that replaces traditional contact solvers in simulators to predict future robot states.

EasyWAM is an open‑source, modular codebase for training and evaluating World Action Models (robotic video‑action predictors). It supports multiple backbones, LoRA fine‑tuning, FlashAttention, DeepSpeed, and provides ready‑to‑run recipes for benchmarks such as LIBERO, RoboDojo, and RoboCasa365.