SOTA · embodied

Embodied AI

Real-robot manipulation, navigation, VLA and whole-body control

Embodied AI is the interface between AI and the physical world, and the cell with the largest current capability gap: what language can already write down, a robot arm often cannot do. Its relation to AGI lies in irreversibility - a physical robot has no save file, one failure can damage an object, a machine or a person, and that forces this domain to solve everything an agent can dodge in pure software: real-time behaviour, sensor noise, contact dynamics, safety boundaries. This domain has two real rulers, which puts embodied ahead of the games domain. On the physical side, the RoboDojo real-robot board (2026-09-22) is led by Pi-05 with an average of 22.9 (arxX5 27.7 / piper 32.0 / piperX 9.1), followed by InternVLA-A1 at 12.0, GalaxeaVLA (G0) at 9.0 and Xiaomi-Robotics-0 at 7.9. In simulation, the RoboDojo sim board is led by Dexmal's DM0.5 at an average of 24.9 (longHorizon 33.7, precision 24.82, generalizationStd 23.49), then GalaxeaVLA (G0.5) at 20.23 and Xiaomi-Robotics-1 at 20.07. Reading the two boards against each other is itself the single most important piece of intelligence in this domain: the simulation leader scores in roughly the same range as the real-robot leader, but the ranking is completely reshuffled, which says sim-to-real transfer remains the main source of loss in the field.

RULERCALVIN, LIBERO, Open X-Embodiment, real robots

The ladder

Embodied AI

Genesis World: The Unified Multi-Physics Simulation Platform for Physical AI

Genesis AI's open simulation platform for physical AI: one Pythonic API over a multi-physics engine (rigid, FEM, MPM, PBD, SPH, Stable Fluid, IPC), the Nyx path tracer, and Quadrants, a cross-platform Python-to-GPU compiler. It is built to be the evaluation engine for robot foundation models rather than a data factory: 30,000 parallel environments at 43M FPS, and under a zero-shot real-to-sim protocol, sim-vs-real evaluation agreement of Pearson 0.8996 with mean maximum rank violation 0.0166. Apache 2.0, 29.9k stars.

30kGitHub stars

Site Repo

Embodied AI

Dexbotic: A One-Stop VLA Toolbox for Pretraining, Fine-Tuning, Inference, and Evaluation

A one-stop VLA toolbox from Dexmal: the unified DexData format spans pretraining, fine-tuning (full/LoRA/RL), inference, and evaluation, supporting mainstream models (π0, CogACT, OFT, MemVLA, GR00T N1) plus the in-house dual-expert DM0; DB-pretraining brings consistent gains across five simulation benchmarks, with 62% average success on RoboChallenge Table30 real-robot evaluation.

1.4kGitHub stars

Site Repo

Embodied AI

ASPIRE: Agentic /Skills Discovery for Robotics

ASPIRE is the continual-learning system NVIDIA open-sourced with the University of Michigan, UIUC, Berkeley and CMU (paper arXiv:2607.00272, Apache-2.0). No gradient descent: a coding agent repeatedly writes, runs and repairs robot control programs (code-as-policy) inside a closed-loop execution engine, promotes only verified fixes into a reusable skill library, and uses evolutionary search to escape single-trajectory patching. Driven by Claude Opus 5.5 it averages 99% success across all 80 LIBERO-Pro tasks and 100% on representative Robosuite and BEHAVIOR-1K tasks; perturbed manipulation improves by up to 77%, and unseen long-horizon tasks reach 31% zero-shot against a 4% baseline. On hardware, skills discovered in simulation transfer across embodiments to a real dual-arm station, and online learning even teaches the robot to solve a Rubik's Cube.

884GitHub stars

Site Repo

Embodied AI

Bimo: a 45 cm Open-Source Hip-Head Biped Trained in Isaac Lab and Distilled onto an RP2040

Bimo is a 45 cm, roughly 1.6 kg hip-head biped kit: eight STS-3215 bus servos, a BNO08x 9-DoF IMU, four VL53L0X rangefinders, two 180-degree cameras, and a custom RP2040 board closing a 20 Hz control loop. The repository publishes the Python control API (1023 lines), three MCU firmware builds (1349 lines), a ROS2 wrapper (1431 lines) and an Isaac Lab training environment (874 lines: six reward terms, dense domain randomization, and a system-identified STS3215Actuator with 3.113 Hz bandwidth, directional gear backlash and a 5 ms bus delay), plus a 26MB Bimo.usd model. The CPG gait's 104 Fourier coefficients and per-joint amplitude gains are hardcoded in both Python and C, and a [256,128,64] PPO teacher distills into a [64,32] student that goes through onnx2c into the firmware, so the robot walks untethered on the MCU. All code is Apache-2.0, but the CAD and electronics the README promises are still marked coming soon: no STL, STEP, schematic, gerbers or BOM exist in the working tree or in git history, and neither release ships pretrained weights (zero assets) or a DIY assembly manual. Currently v1.1.0 with 198 stars, in pre-order status.

198GitHub stars

Site Repo

3D & World Models

Text-to-CAD: the open-source plugin and skill suite that gives coding agents CAD superpowers

text-to-cad turns parametric CAD into a local agent skill: plain language or an image becomes a STEP model with STL/3MF/GLB exports, plus DFM review, dimensioned drawings, URDF/SRDF/SDF robot descriptions and 3D printing, sheet metal and CNC integrations. Kernel is build123d on OCP; the shape is twelve skills plus the cadgen engine and a local CAD Viewer. MIT, about 20.6k stars.

21kGitHub stars

Site Repo

Latest demos

1112

0:12

Papers & deep reads

392

2026-11-02MABEL: An Open Anthropomorphic Mobile Bimanual Platform for Teleoperation and LearningAs robot learning moves beyond the tabletop, many tasks require mobility, adjustable reach, active perception, and bimanual dexterity. Commercial platforms combining these capabilities are costly, while affordable open-source alternatives cover only a subset and require software integration. We present MABEL, an open-source anthropomorphic mobile bimanual robot with 56 degrees of freedom (DOF) and a US$8.1k parts cost. It combines two 7-DOF arms with 17-DOF hands, a pitching torso, a lift, a 3-DOF neck, and a holonomic base. Back-drivable arms and torso support compliant interaction, with fingertip reach from floor level to 2.23 m. The head and torso roughly triple modeled camera coverage relative to a torso-fixed camera. Apple Vision Pro head and hand tracking enables whole-body teleoperation without handheld controllers, with 137 ms mean glass-to-glass stereo video latency over the robot's Wi-Fi hotspot. We demonstrate an end-to-end imitation learning pipeline, from teleoperated data collection to policy deployment, using a shared command interface and whole-body controller.

2026-11-02D&A: Decoupling Description and Assessment Enables VLMs as Zero-Shot Robotic Reward ModelsVision-Language Models (VLMs) hold promise for zero-shot video task-progress estimation in robot learning, yet existing frameworks frequently suffer from confirmation bias, hallucinating progress in failed executions due to implicit success assumptions or relative temporal ranking designs. Naive prompting of frontier models collapses into a binary success detector that fails to capture nuanced, non-linear progress and failures. While post-training approaches can mitigate this through explicit training, they demand expensive real-world robotics data and require computationally intensive post-training for every individual base model. To resolve these limitations, we introduce D&A (Describe & Assess): our approach decouples progress estimation into an objective, task-grounded textual description of scene dynamics followed by an assessment phase using explicit rubrics. By compelling the VLM to synthesize an explicit, text-based definition of absolute success before scoring, the framework anchors evaluations to an external, global criterion rather than the video's internal timeline. Empirical results demonstrate that our method calibrates zero-shot estimates and mitigates confirmation bias, providing fine-grained, partial progress metrics that accurately penalize failed executions while correctly interpolating successful runs. Being purely prompt-based, D&A generalizes across tasks, embodiments, and benchmarks---matching or surpassing post-trained reward models---and its reward estimates enable efficient downstream reinforcement learning.

2026-11-02RoboColosseum: Evaluating Physical Intelligence at Scale In-The-WildGeneral-purpose robot policies must work across robots, tasks, and environments, yet the variation that makes physical trials informative also makes them difficult to reproduce. We introduce RoboColosseum, a distributed evaluation framework that combines unified policy serving and standardized protocols across a network of locally operated robots in diverse environments. Our initial study evaluates five vision-language-action (VLA) policies and three frontier AI agents across four embodiments (SO-101, Franka, Bimanual-YAM, and G1 Humanoid). The Open Track compares models without task-specific adaptation in blinded, pairwise trials under matched local conditions. The Fine-tune Track measures how well VLA policies adapt to 20 tasks across all four embodiments when each policy receives the same 100 demonstrations per task. Across 5,699 pairwise Open Track comparisons at 124 sites, frontier agents rival specialized VLA policies, while all systems struggle to complete open-world tasks reliably. After fine-tuning, GR00T N1.7 leads on SO-101 and pi0.5 leads on Franka and Bimanual-YAM. Policy rankings show weak agreement across embodiment-specific task suites (Kendall W = 0.24). Open Track rankings poorly predict relative performance after fine-tuning, with substantial rank reversals on all three embodiments shared between tracks. Measuring partial progress reveals differences that task success alone misses. On the G1 Humanoid, no policy completes any of the five tasks, yet average progress ranges from 13.3% to 23.2% across policies. Open Track rankings stabilize as trials accumulate, and active matchmaking achieves comparable ranking agreement with substantially fewer physical trials than uniform sampling. These results show how distributed physical evaluation can reveal where robot policies generalize, how well they adapt to new tasks, and how far they progress on tasks they fail to complete.

2026-11-01From RT-1 to MolmoAct 2: A Brief History of Vision-Language-Action ModelsA complete history of VLAs from SayCan and RT-1 through RT-2, RT-X, Octo, OpenVLA, pi0, SmolVLA and MolmoAct 2, including the author's action-discretization experiment.

2026-10-30Movens-Zero: Scaling Humanoid Motion Foundation Models with Large-Scale Human VideosMovens-Zero builds five million executable humanoid motions from Internet human videos, unifies text, video, speech, and music conditioning, and drives Unitree G1 in real time through chunked closed-loop control.

2026-10-27METATRON: A Promptable Whole-Body Execution Layer for HumanoidsHumanoid controllers often require task-specific training despite sharing the need for balance and whole-body coordination. We present Metatron, a promptable whole-body execution layer combining a generative base policy with a programmable skill layer. The base policy receives partial goals, collectively called a prompt, that can combine timed body-part targets, velocity commands, and style cues. It produces diverse whole-body motions consistent with these goals. Training combines motion capture, synthetic recovery trajectories, and retargeted scene-interaction motions. After training, the policy is frozen. Stateful programs called skills express task intent as prompts at each control step, drawing rough geometric sketches that the policy turns into coordinated movement. Skills are ordinary code, refined by coding agents that analyze motion data and fit compact rules without task-specific policy training. On the physical Unitree G1, one frozen checkpoint supports walking, recovery, teleoperated object pickup, and sitting on furniture. Comparative evaluations show higher reaching success and push robustness than the tested reference-tracking pipelines. The same recipe produces climbing and acrobatics on a simulated character. Metatron enables coding agents to turn learned motor capabilities into new skills through code, without further policy training.

2026-10-26Learning Multi-Phase Robot Loco-Manipulation via Phase-Conditioned Policy OptimizationComplex whole-body loco-manipulation requires executing diverse maneuvers, such as crouching, grasping, and locomotion, within a single long-horizon task. While performing simple pick-and-place behaviors is well-established, scaling to complex, long-horizon tasks remains difficult. Currently, this challenge is addressed by relying on human motion data and correspondingly humanoid platforms, which limits platform independence and introduces substantial data processing overhead. Instead, we extend proximal policy optimization (PPO) to train a single policy for multi-phase tasks across diverse robotic hardware, leveraging prior task structure to shape behavior. Our method adapts PPO using a phase-conditioned critic, phase-gated reward functions, and a multi-phase reset curriculum to guide exploration. To support varied hardware, we employ a decoupled hierarchical control architecture that separates lower-level balancing dynamics from high-level phase execution. We demonstrate multi-stage loco-manipulation skills on hardware on both the Unitree G1 and a Boston Dynamics Spot equipped with a manipulator arm, and validate our design choices through ablations. Our approach enables the acquisition of complex loco-manipulation behaviors purely through environment interaction while directly incorporating high-level task structure into policy learning.

2026-10-25InteractiveWorld: A Verifiable Human-Centric Environment for Building Better Interactive MLLM AssistantsPeople have long dreamed of an assistant that is always there to guide them and remind what they forget, from new dishes to lost keys. Powered by multimodal large language models (MLLMs), egocentric AI assistants on wearable devices are bringing this dream within reach: they see what the user sees and give instructions for user's physical work when necessary. However, testing such assistance is challenging and usually requires real people. The expensive and rare real interaction data leaves a key question unanswered: Can today's MLLMs truly help people in physical world? To answer this question, we propose InteractiveWorld, the first verifiable environment for closed-loop interaction between human and AI assistant. Unlike existing robot-centric simulators, InteractiveWorld lets the assistant guide a human agent who observes the scene, understands instructions, and acts independently. It thus realistically simulates how AI assistant collaborates with human: the assistant instructs, the human acts, and the environment executes actions and updates what they see next. Building on it, we introduce InteractiveBench, the first dynamic benchmark for egocentric AI assistants, with 120 multi-turn tasks across six categories. In each task, the assistant guides the human in real time, and every instruction shapes the outcome, which is scored by verifiable, state-based rewards. Experiments with 12 frontier MLLMs show they are still far from reliable assistants. Even the strongest model, Claude Fable 5.1, achieves at most 20% and 25% on long-horizon Guidance and Planning tasks, respectively. Overall, our work offers a verifiable platform and a rigorous evaluation set, laying a solid foundation for research on interactive MLLM assistants that truly help people in the physical world.

2026-10-25MorphMimic: Morphology-Conditioned World Model and Policy Sharing Across HumanoidsMorphMimic conditions a shared differentiable world model and tracking policy on a 125-value morphology descriptor, enabling three humanoids to learn from 128,786 common motion clips and share one controller.

2026-10-23RynnUnify: Towards Unified Embodied BrainWe introduce RynnUnify, a 16B unified embodied model that combines strong embodied understanding with long-horizon visual planning through interleaved video--language reasoning. With capabilities spanning spatial reasoning, physical prediction, and grounding, RynnUnify connects an understanding of the physical world with visual anticipation of how it evolves across successive subtasks. Building on these capabilities, we further develop an agentic extension through reinforcement learning to selectively use video predictions for reasoning about unobserved states and interaction outcomes. We also develop a world-action extension of the full RynnUnify model, translating its understanding and prediction capabilities into robot control. To evaluate these capabilities, we introduce Rynn-VP and Rynn-GEU for long-horizon visual planning and anticipatory reasoning. Experiments show that RynnUnify achieves the highest average embodied understanding score among the evaluated unified multimodal models and improves the Rynn-VP overall score by 9.75 points over Cosmos 3 Nano. Its agentic extension improves the Rynn-GEU average by 6.0 points over direct reasoning, while the world-action extension achieves 51.5\% success on 20 selected composite RoboCasa365 tasks and 46.7\% average success across three real-robot tasks. Together, these results establish RynnUnify as a shared foundation for embodied understanding, visual planning, and robot control, taking a step toward a unified embodied brain.

2026-10-23Mundo: From Seeing a Place to Simulating What Happens ThereA few photos become a world you can explore, edit and simulate: a 3D proxy plus a code scene program drive simulation while a video model renders it, supporting new views and Real-to-Sim robotics.

2026-10-22FLORA: An Open Model Suite for Scaling Robotics Data EnginesThe life-cycle of robotics foundation models crucially depends on data labels. Outcome reward labels are needed for monitoring real-world deployments. Process reward labels are needed during RL post-training. Timestamped captions are needed for filtering and balancing pretraining data. We present Flora, an industrial-quality data engine used for training GR00T models, that simultaneously covers these capabilities. Flora achieves SOTA scores on the captioning benchmark WGO-Bench and reward benchmarks RoboReward and Robometer. On internal benchmarks it is competitive with GPT-6 Astra while beating other frontier models. The model is substantially cheaper than API-based alternatives, roughly 125 times cheaper than GPT-6 Astra. Flora is battle tested at scale, labeling 200,000 hours of data per day across 512 GB300s. By open-sourcing our data engine, we enable researchers to scale custom robotics datasets and explore emerging applications such as physical auto-research and human-to-robot transfer.

2026-10-22Humanoid Mobility via Attention-based ContactsHumanoid mobility over complex terrain requires accurate foot placement and coordinated whole-body motion capable of robust adaptation. Reinforcement learning (RL) policies trained without motion references generalize well across terrains, yet eliciting effective whole-body coordination through reward design and exploration alone remains challenging. In contrast, reference-based methods yield natural motions but struggle to adapt beyond the predefined trajectories in their training data. We introduce a learning framework that bridges these two approaches by using motion data strictly as a training prior. An explicit, self-supervised, attention-based contact planner predicts future body-environment contacts, while a contact-conditioned tracker translates these plans into coordinated whole-body motion. Together, they establish contact as an interface between sensory observations and control actions. This abstraction allows the policy to internalize the contact-aware structure of locomotion rather than directly replicating kinematics. Consequently, a single policy can exploit data-driven motion priors for natural whole-body coordination while retaining the flexibility to discover environment-dependent behaviors through reinforcement learning. Building on the learned contact planner and tracker, a subsequent fine-tuning stage bootstraps their acquired capabilities to explore environment variations beyond those covered by the original motion data. Operating entirely from onboard observations without any reference inputs at deployment, this framework enables versatile humanoid mobility in challenging environments.

2026-10-22See What I Need, Dream What I Want: A Navigation Foundation Model on a Self-Constructed Representation SpaceRepresentNav learns which features a navigation policy needs across three frozen vision backbones, predicts four future states in that same latent space, and emits eight waypoint poses; simulated benchmarks, latent probes, and physical-robot trials test whether task-shaped seeing and dreaming improve control.

2026-10-21DONIC: Generalist Dexterous Bimanual Tracking Control from Human DemonstrationsDespite recent progress in dexterous reference tracking, existing models focus primarily on single-hand manipulation and a limited range of interactions, leaving the bimanual dexterous behaviors common in real life unaddressed. The scarcity of contact-rich bimanual manipulation data and the increased complexity of coordinating contacts make developing a generalist bimanual tracker particularly challenging. Therefore, we present 1) DONIC, the first generalist bimanual tracker jointly trained across operation categories and human-motion sources, together with 2) BiHOI, a multi-view dataset of 1,575 sequences spanning 229 objects and emphasizing sustained bimanual interaction, where reconstructed interactions are retargeted into robot references. DONIC combines proximal policy optimization with behavioral cloning in a feedback loop that recycles verified successful executions into training demonstrations. An object-centered per-hand reward maintains object-motion feedback while supervising each hand according to its reference contact role. Clip-Centered Advantages remove reference-dependent advantage offsets while preserving within-trajectory differences. Together, these designs enable one single policy to learn coordinated manipulation across heterogeneous references. Without finetuning, a single checkpoint achieves 86.79% success on unseen trajectories of training objects and 57.32% on held-out objects, surpassing the state-of-the-art bimanual tracking baselines by 77.36 and 45.73 percentage points.

2026-10-21Agentic World Action ModelsRecent world action models provide visual foresight and future prediction for robot manipulation, but lack explicit mechanisms for compositional skill reuse. Agentic robotics combines code-based planning and reasoning with reusable skills, but often lacks predictive grounding in physical dynamics. We introduce AW-0, an agentic world action model framework that integrates code-based planning with joint video and action prediction. We jointly model code, future video, and robot actions as three coupled generative streams within a unified mixture-of-transformers architecture. Within this architecture a code expert generates parameterized skill routines and control logic while video and action experts generate future visual states and low-level motor commands. Code serves two complementary roles: it structures multi-step prediction and control within the world model and represents skills as reusable programs for subsequent agentic execution. We further introduce an automated curation pipeline that labels robot and human video trajectories with aligned programs and consolidates recurring routines into reusable skill libraries. Evaluations on RoboTwin 2.0 and physical bimanual manipulation tasks show improved task performance over VLA and WAM baselines, alongside gains in language following, semantic reasoning and robustness, demonstrating that agentic code enables connection between planning, prediction and execution within WAMs.

2026-10-20T-WLA: Bridging Slow Vision and Fast TouchT-WLA shares a history-aware visual memory between a slow vision-language-action policy and a fast tactile residual editor. The visual branch produces an action chunk and memory at 39.7 Hz; the lightweight tactile editor reuses them to correct only unexecuted actions at 119.1 Hz. It improves tactile-conditioned hand-motion prediction and contact-rich manipulation, reaching 50.0% average success on four real-robot tasks with 30 demonstrations each.

2026-10-20SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World ModelSuperMimic fits one differentiable world model from simulator transitions and uses short imagined rollouts to train a general humanoid tracker. It surpasses SONIC in three hours on one H100, then reaches zero failures on 129,514 motions after 9.8 hours on two H100s.

2026-10-20Dream Labs Comes Out of Stealth With $90M Seed: World Action Model Dream-1 Drives 27 Unseen Robots Zero-Shot at $3 an HourDream Labs emerges from stealth with $90M in seed funding and Dream-1, a World Action Model that drives 27 robots it was never trained on over a public API at $3 an hour, about $0.0013 per action.

2026-10-18Vision as Proprioception: Estimating Humanoid Robot Pose from VideoHumanoidPose combines frozen human-pose features with robot kinematics to recover full-body joint configurations and world-space motion from monocular video across embodiments.

Open source

NVlabs/ASPIREApache-2.0

ASPIRE: Agentic /Skills Discovery for Robotics

Code-as-PolicySkill Library

Python884784

amazon-far/RPLApache-2.0

RPL: Learning Robust Humanoid Perceptive Locomotion on Challenging Terrains

CoRL 2026 open-source release from Amazon FAR with CMU, Stanford and UC Berkeley: per-terrain multi-actor PPO experts trained on privileged height maps are distilled into a dual-depth-camera vision transformer student, robustified by velocity-gated depth feature scaling and random side masking, and backed by a multi-depth ray-casting pipeline 5x faster than IsaacSim Warp. On a Unitree G1 the single policy walks bidirectionally with a 2 kg payload over 20 degree slopes, 22/25/30 cm stairs and stepping stones with 60 cm gaps.

HumanoidPerceptive Locomotion

Python1530

earthtojake/text-to-cadMIT

Text-to-CAD: the open-source plugin and skill suite that gives coding agents CAD superpowers

CADAgent Skills

Python21k2.0k100

gentlefress/Omega-0NOASSERTION

Omega-0 Official Code: Training, Inference and Unitree G1 Deployment for a Latent Predictive World-Action Model

Official implementation (MIT) of the omega-0 paper: a world-action model with Qwen3-VL conditioning, a frozen ViT-g frame encoder and Wan2.2 VAE latent supervision, whose 6-block three-stream predictor drives a flow-matching diffusion head over 66-dim whole-body actions (30-step chunks, real-time chunking). The repo also ships the Unitree G1 robot stack: versioned msgpack protocol, module runtime, SONIC whole-body adapter, Pico teleoperation collection and HDF5 recording. The 7.42 GB pretrained checkpoint landed on 2026-10-12, completing a release whose 4,828-episode omega-HOME dataset had been public since September.

World-action modelsHumanoid

Python5632

kevinzakka/mjbatchApache-2.0

mjbatch: Batched MuJoCo Simulation on a CPU Thread Pool

A small Python + C++ library from Kevin Zakka that steps thousands of MuJoCo simulations in parallel on the CPU through a single Batch object: C++ thread pool execution with the GIL released, bind() giving live array views across the whole batch, and expand() for per-simulation model parameters. Memory scales with thread count rather than simulation count, so 4096 simulations stay under 256MB; on a 24-thread machine with a Unitree G1 scene it measured 10.3x over a serial mj_step loop, reaching 676,019 sim-substeps per second with ten substeps batched per call. The repo ships six self-contained solver examples spanning iLQR, predictive-sampling MPC, PPO reinforcement learning, CEM hardware co-design and damped Gauss-Newton system identification, including a Go1 joystick controller that learns to walk in under a minute on a five-year-old M1 laptop.

SimulationMuJoCo

Python2802

mekion/the-bimo-projectApache-2.0

Bimo: a 45 cm Open-Source Hip-Head Biped Trained in Isaac Lab and Distilled onto an RP2040

Bipedal robotsHumanoid

Python19887

Genesis-Embodied-AI/genesis-worldApache-2.0

Genesis World: The Unified Multi-Physics Simulation Platform for Physical AI

SimulationPhysics Engine

Python30k2.9k224

reubenstr/ElectroPup

ElectroPup: a 12-DoF 3D-Printed Open-Source Quadruped

ElectroPup is a 4.5 kg, 12-DoF DIY quadruped: twelve MG4010E-i10v3 CAN BLDC actuators split into four per-leg CAN segments, a Raspberry Pi 5 for compute, and a single ~7,700-line pure-Python service with no ROS in the stack. The repository publishes OnShape CAD, two KiCad v8 boards, a bill of materials, measured weights and power draw (21.6 W suspended, 33.8 W standing), plus MuJoCo simulation, a React Native/Expo browser UI, four Bezier-foot gaits, and a full build guide. The author also documents known gaps: no license file, no contact sensing, and a simulation model without approximate masses or inertias.

Quadruped robotsOpen Source

Python3542

molanocortes/takto-oneNOASSERTION

Takto One: Open-Source Hand Exoskeleton

Open-source hand exoskeleton: 12 joint encoders, 8 tendon-driven motors, on-device Teensy control, and synchronized capture/EMG for motion capture, teleoperation and haptics research.

EMGTeensy

JavaScript1000

dexmal/dexboticMIT

Dexbotic: A One-Stop VLA Toolbox for Pretraining, Fine-Tuning, Inference, and Evaluation

VLAManipulation

Python1.4k1814

lightorigins/LightNav-0Apache-2.0

LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation

A compact generalist navigation model from Light Origins: Qwen3-VL-4B backbone + dual-channel pointing + RVQ action tokens. One checkpoint covers instruction following, object navigation and visual tracking, transferring zero-shot across humanoid/quadruped/wheeled/aerial robots. Trained entirely in simulation, Apache-2.0.

LightNav-0LightOrigins

Python6120

mrzuang/GRIT_teleop_deployMIT

GRIT: Whole-Body Motion-Tracking Control and VR Teleoperation Deployment for Unitree G1

GRIT is a deployment framework for whole-body motion-tracking control on the Unitree G1 humanoid: ONNX policy, 50 Hz Python inference runtime, MuJoCo sim2sim, live PICO VR full-body teleoperation, and a native SDK2 hardware bridge with stale-command watchdog — a complete low-barrier pipeline for whole-body control and teleoperation data collection.

Unitree G1MuJoCo

Python2700

Assets you can use now

No installable skill wired to this domain yet.

Verticals using it

Boundaries & failure modes

The readings in this domain have to be interpreted with physical intuition, or they will mislead you. One, the absolute level is still low: the real-robot leader Pi-05 averages 22.9, meaning under RoboDojo's task protocol success is below a quarter, and second place at 12.0 falls off a cliff - this domain has not yet formed the dense first tier that mature fields have, so a single-point lead may come from task bias rather than general capability. Two, per-platform variance is enormous: Pi-05 scores 32.0 on piper but only 9.1 on piperX, the same model differing more than threefold across hardware, while vendor releases usually quote only the best platform. Per-platform readings are therefore printed in this domain, and citing the average alone is not allowed. Three, simulation and real-robot rankings swap: DM0.5 leads the sim board at 24.9 but is not in the top tier on real hardware, and Pi-05 is the reverse, and that inconsistency itself says simulation scores cannot be extrapolated to real robots, so any claim of "superhuman in simulation" needs separate verification. Four, long horizon and memory are the weakest axes: the longHorizon sub-scores on the sim board sit mostly between 23 and 44 while precision is higher, which says current VLA systems are good at short-range dexterous manipulation and bad at multi-step procedures. Five, generalisation to random perturbation barely exists: generalizationRand runs from 1.57 to 11.44, the lowest band across all dimensions, and real-world changes in lighting, material and pose are exactly that kind of perturbation. Six, cost and data: real-robot data cannot be scraped from the internet, one teleoperated demonstration costs human minutes, and that is the structural reason this domain advances more slowly than language, not a lack of effort.

As an Amazon Associate, we earn from qualifying purchases.

The ladder

Genesis World: The Unified Multi-Physics Simulation Platform for Physical AI

Dexbotic: A One-Stop VLA Toolbox for Pretraining, Fine-Tuning, Inference, and Evaluation

ASPIRE: Agentic /Skills Discovery for Robotics

Bimo: a 45 cm Open-Source Hip-Head Biped Trained in Isaac Lab and Distilled onto an RP2040

Text-to-CAD: the open-source plugin and skill suite that gives coding agents CAD superpowers

Latest demos

LinkerBot OYI: A 1.1 m, 21-DoF Humanoid for Education

Two Models Attempt a Multi-Step Desk Organization Task

frax.ik: Microsecond Humanoid Retargeting on CPU

DISCO Distills Continuous-Action Tree Search into Policies

Liber-0 turns human experience into zero-shot robot manipulation

Perceptron Egocentric 1.0 Annotation API

Perceptron Cuts Robot Video Annotation Cost to One-Eighth

Astra Edits Four Weights to Lift VLA Success

PNDbotics Assembles a Humanoid Like a LEGO Set

FLoRe Learns Force-Reactive Humanoid Loco-Manipulation

Galaxea G0.5 VLA Joins LeRobot

D&A Turns Any VLM Into a Robot Reward Model

Pixel: A Low-Cost Full-Stack Humanoid for Mobile Manipulation

PhysiClaw Gives AI Agents a Physical iPhone Loop

GPT-6 Astra Completes All 20 Spot Cleanup Trials

Zero-Shot Mobile Manipulation on Dexmate

A3 Ultra navigates floors to tidy rooms

EgoOne Lite: A Wearable Egocentric Data Collection Rig

G1 Goalkeeper Turns Live Vision Into Saves

InflatableBots Bring Large-Scale Haptics to VR

Darwin Builds Robot Technicians for Data Centers

1X NEO Head Prototype Navigates an Apple TV

RAVEN Stores Long-Horizon Robot Memories as Visual Embeddings

OM-1 Learns Stair Grasping from Human Demonstrations

Papers & deep reads

Open source

ASPIRE: Agentic /Skills Discovery for Robotics

RPL: Learning Robust Humanoid Perceptive Locomotion on Challenging Terrains

Text-to-CAD: the open-source plugin and skill suite that gives coding agents CAD superpowers

Omega-0 Official Code: Training, Inference and Unitree G1 Deployment for a Latent Predictive World-Action Model

mjbatch: Batched MuJoCo Simulation on a CPU Thread Pool

Bimo: a 45 cm Open-Source Hip-Head Biped Trained in Isaac Lab and Distilled onto an RP2040

Genesis World: The Unified Multi-Physics Simulation Platform for Physical AI

ElectroPup: a 12-DoF 3D-Printed Open-Source Quadruped

Takto One: Open-Source Hand Exoskeleton

Dexbotic: A One-Stop VLA Toolbox for Pretraining, Fine-Tuning, Inference, and Evaluation

LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation

GRIT: Whole-Body Motion-Tracking Control and VR Teleoperation Deployment for Unitree G1

Assets you can use now

Verticals using it

Boundaries & failure modes