Robot brains

• 1X World ModelWorld model · Commercial

  1X Technologies' generative world model plus VLA stack powering NEO, the consumer-home humanoid shipping via preorder. The world model enables learning by predicting outcomes versus exhaustive pre-programming. Combined with VLA control for action generation. Important verified-vs-claimed flag: NEO home capabilities are teleoperated-assisted; the autonomy-vs-teleoperation distinction is the load-bearing consumer verification point.

• AeroVironment AVACOREFramework · Production

  AeroVironment's open autonomy software architecture: a behavior-tree-based, airframe-decoupled stack paired with the Autonomy Retrofit Kit (ARK) edge-compute payload and SPOTR-Edge perception. Designed to be cross-platform; it underpins the Red Dragon one-way-attack UAS and is a named retrofit for the Puma family.

• Anduril LatticeOS-layer · Production

  Anduril's AI command-and-control and mission-autonomy software platform: it fuses sensors and effectors into a common operating picture and autonomously tasks teams of unmanned systems under a single operator. An integration/OS layer for autonomous systems rather than a single-platform autopilot.

• Apollo ADFM (Baidu)Foundation model · Commercial

  Baidu's Apollo autonomous-driving platform, rebuilt around Apollo ADFM (Autonomous Driving Foundation Model) in Apollo Open Platform 10.0 (Dec 2024), described by Baidu as the world's first foundation model supporting Level 4 autonomous driving. Powers the Apollo Go robotaxi fleet (including the 6th-gen RT6) and is credited with enabling driverless deployment in new cities within months.

• Atlas-GeminiFoundation model · Pilot

  The hybrid brain powering Boston Dynamics' all-electric Atlas. Combines Boston Dynamics' Atlas control stack (including Orbit skill-sharing) with Google DeepMind's Gemini Robotics foundation models for higher-level reasoning and learning. Partnership announced at CES January 2026; Atlas 2026 production is committed to Hyundai plus DeepMind, with the Robotics Metaplant Application Center (RMAC) at Hyundai Metaplant America serving as a supervised-training data factory.

• Aurora DriverFoundation model · Commercial

  Aurora's SAE Level 4 autonomous-driving system for trucks: the common autonomy platform (sensing, including long-range lidar, plus redundant compute and the driving software) that operates Class 8 trucks driverlessly. Integrated across PACCAR, International (Navistar), and Volvo truck platforms.

• Avride ADSFoundation model · Pilot

  Avride's in-house automated driving system (ADS), developed by the core team behind Yandex's former self-driving group. A single proprietary autonomy stack (lidar, camera, ultrasonic) powering both Avride's robotaxis (fully-electric Hyundai IONIQ 5, on the Uber platform in Dallas and Austin) and its sidewalk delivery robots across US cities.

• Bot Auto Autonomous Driving SystemFoundation model · Commercial

  Bot Auto's Level 4 autonomous-driving system: a transformer-based, end-to-end ('foundation-to-all') neural network using camera, radar, and lidar with redundant power/steering/braking. It has no separate product brand name (referred to generically as Bot Auto's autonomous driving system). Operated Bot Auto's first fully humanless commercial truckload in April 2026.

• CarbonFoundation model · Commercial

  Sanctuary AI's Carbon cognitive AI control system. A cognitive hybrid that translates natural language into precise physical actions, with emphasis on human-like hand dexterity. Powers Phoenix, Sanctuary's humanoid known for fine manipulation tasks such as buttoning shirts and handling fragile lab tools. Framed by Sanctuary as 'a brain for work.'

• DYNA-1Foundation model · Commercial

  Dyna Robotics' DYNA-1 is a dexterous robot foundation model for sustained autonomous manipulation, launched April 29, 2025 from Redwood City, California. It runs on a pair of stationary dual robotic arms and is trained with imitation learning plus reinforcement learning against a proprietary custom reward model that lets the system self-recover from errors and generate its own training data, which Dyna positions as its core technical differentiator. The company describes it as a single-weight, general-purpose vision-language-action model but has not disclosed its parameter count or published a technical paper, and its weights and code are closed. Dyna frames DYNA-1 as the first dexterous robot foundation model deployed in commercial settings, a company superlative the registry records as a claim rather than a verified industry first. Its disclosed deployment environments are hotels, restaurants, laundromats, and gyms, with robots reported running about sixteen hours a day after six months, but no specific customer, site, or unit count is named, so the commercial deployment is recorded as demonstrated and early or pilot stage rather than verified at scale. Its headline results are company-reported demonstrations: more than eight hundred napkins folded across twenty-four continuous hours with zero human intervention at a 99.4 percent success rate, at roughly 60 percent of human throughput, and a 99.9 percent success rate self-reported via Dyna's X account at the three-day HITEC 2025 trade-show demo. Dyna Robotics was founded in 2025 by Lindon Gao, the chief executive, and York Yang, who previously built the smart-cart company Caper AI that Instacart acquired in 2021, together with co-founder Jason Ma, a former research scientist at Google DeepMind. It raised a 23.5 million dollar seed round in March 2025 led by CRV and First Round Capital, and a 120 million dollar Series A in September 2025 led by RoboStrategy, CRV, and First Round Capital with participation from NVentures, the Amazon Industrial Innovation Fund, Salesforce Ventures, Samsung Next, and LG Technology Ventures; a post-money valuation above 600 million dollars is reported but not company-confirmed. No successor model had been announced as of June 2026.

• Einride DriverFoundation model · Commercial

  Einride's proprietary Autonomous Drive Stack (the Einride Driver), the Level 4 system that operates the cabless electric Einride Pod on public roads under remote supervision (a Remote Pod Operator monitors). Distinct from Saga, Einride's AI-powered freight operating system (fleet/route layer). Runs full-time daily commercial autonomous operations (e.g. for GE Appliances) and completed the world-first cabless autonomous border crossing (Norway-Sweden, 2025).

• Embark DriverFoundation model · Research

  Embark's Level-4 autonomous-driving stack for long-haul trucks. Reached test/pilot operation with safety drivers but never observer-free commercial deployment. Defunct: Embark wound down in 2023 and its IP was acquired by Applied Intuition.

• Field Foundation Models (FFMs)Foundation model · Commercial

  Field AI, a US startup led by founder and chief executive Ali Agha, builds Field Foundation Models, which it describes as physics-first, embodiment-agnostic autonomy models for GPS- and map-denied unstructured environments that let robots navigate without maps, GPS, or pre-defined trajectories. The models are proprietary, with no published model card or paper, the technical claims living on company and PR pages and in patents. The registry records the brain at commercial maturity on the strength of the strongest deployed-wiring story among the newly added emerging brains, though a customer-anonymized one: the models run across humanoids, quadrupeds, and wheeled and legged platforms in construction, energy and oil and gas, mining, logistics and urban delivery, inspection, and federal work, with the company citing deployments across hundreds of complex real-world industrial environments and multiple expansion contracts, credible commercial traction but with no named customer or site in primary sources. Field AI has raised $405 million across two rounds, the larger a $314 million oversubscribed round announced August 20, 2025 at a reported valuation around $2 billion, from investors including Bezos Expeditions, Nvidia's NVentures, Khosla, Intel Capital, Temasek, BHP Ventures, and Samsung. It occupies a distinct niche, industrial and field robotics in GPS-denied environments, versus the manipulation- and driving-focused brains, while its model card and paper are absent so all model claims are company-sourced, its deployments are real but anonymized and unverifiable at the named-site level, and its physics-first architectural specifics and precise valuation are not verified.

• Gatik DriverFoundation model · Commercial

  Gatik's L4 autonomous-driving system (3rd-generation) for fixed, repeatable middle-mile routes, paired with the Gatik Arena simulation platform (built on NVIDIA Cosmos) and NVIDIA DRIVE AGX Thor in-vehicle compute. Operates Class 6/7 box trucks driverlessly on commercial routes.

• Gemini RoboticsFoundation model · Research

  Google DeepMind's robotics VLA family. Gemini Robotics builds on Gemini 2.0 with physical actions as an output modality and adds an intermediate reasoning layer for spatial analysis and safety. Gemini Robotics-ER is an embodied-reasoning VLM companion. Gemini Robotics On-Device runs locally, network-independent, and is fine-tunable with 50-100 demonstrations. Gemini Robotics 1.5 introduced 'think before acting'.

• GEN-0 (and GEN-1)Foundation model · Research

  Generalist AI (founded 2024 in San Mateo, California by chief executive Pete Florence, a former DeepMind senior research scientist, with Andy Zeng, also ex-Google DeepMind, and Andrew Barry, ex-Boston Dynamics, and backed by Nvidia) builds embodied foundation models trained directly on raw physical-interaction data. Its GEN-0 model, published in a November 4, 2025 research blog, was tested at one to ten-plus billion parameters on more than 270,000 hours of real manipulation trajectories growing by roughly 10,000 hours a week, and reports a scaling phase transition around 7 billion parameters; a successor, GEN-1, followed in April 2026. The models are proprietary with no open weights. The registry records the brain at research maturity because its capability is demonstrated rather than deployed: it runs across 6-, 7-, and 16-plus-degree-of-freedom semi-humanoid robots in lab and benchmark settings, but there is no named hardware partner, customer, or production deployment, and the robot wiring is in-house and captive. The artifact is real and substantive enough to clear the registrable bar even without peer review, but its benchmark numbers, including a 99-percent task-success claim, the 270,000-hour figure, and the scaling claims, are self-reported and not independently reproduced, there is no arXiv paper or formal model card, the funding amount and valuation are undisclosed, and there is no named robot-platform partnership.

• Genie Operator-1 (GO-1)Foundation model · Pilot

  AgiBot's universal embodied foundation model, Genie Operator-1 (GO-1), launched March 2025. Introduces the Vision-Language-Latent-Action (ViLLA) framework: a Vision-Language Model plus a Mixture-of-Experts (a Latent Planner trained on cross-embodiment and human-operation data, and an Action Expert trained on over a million real robot demonstrations). Supports learning from human video, few-shot generalization, cross-embodiment adaptation, and continuous self-evolution.

• GR00T N1 (Isaac GR00T)Foundation model · Research · Open

  NVIDIA's open humanoid foundation model and the first of its kind to be released openly. A dual-system VLA: System 2 uses NVIDIA-Eagle plus SmolLM-1.7B as the VLM at roughly 10 Hz; System 1 is a diffusion transformer producing real-time motor actions. Jointly trained end-to-end on real-robot trajectories, human videos, and synthetic data. Part of the broader Isaac GR00T platform that includes Isaac Lab, Omniverse, and Cosmos.

• GraspVLA (Galbot)Foundation model · Research

  Galbot's grasping foundation model, GraspVLA (launched Jan 2025): an end-to-end Vision-Language-Action model pre-trained on SynGrasp-1B, a billion-frame synthetic grasping dataset with photorealistic rendering and domain randomization. Enables zero-shot generalization to new grasping tasks without additional training, powering dexterous manipulation on Galbot's robots.

• Grok (xAI)Foundation model · Pilot

  xAI's Grok large language model serves as the System 2 conversational and reasoning layer in Tesla Optimus's dual-brain architecture. Grok handles natural-language understanding and high-level instruction reasoning (the "what should I do next" layer), while Tesla's FSD-derived neural networks handle the System 1 visuomotor layer (the "actually doing it" layer). The same Grok-reasoning-plus-Tesla-compute pattern is reused in the "Digital Optimus" (Macrohard) software-agent project announced March 2026.

• Helix (and Helix-02)Foundation model · Commercial

  Figure AI's onboard dual-system VLA. System 2 is an internet-pretrained VLM (7B params, 7-9 Hz) handling scene and language understanding; System 1 is a fast reactive visuomotor policy (80M params, 200 Hz). Helix-02 (January 27, 2026) extended to full-body control via System 0, a 1 kHz neural prior trained on 1,000+ hours of human motion data that replaced roughly 109,504 lines of hand-engineered C++. The first VLA to control a full humanoid upper body including individual fingers from one set of weights, and the first to run two robots from one set of weights.

• IntEngine (IntBot)Foundation model · Pilot

  IntBot's proprietary multimodal, multi-loop social-intelligence system (IntEngine, also styled its General Social Intelligence Engine). Fuses vision, audio, and language in real time to coordinate speech, facial expression, and gesture, letting a robot perceive social context and decide when and how to engage in open public settings. Hardware-agnostic across multiple humanoid and service platforms; edge-deployed on NVIDIA Cosmos Reason. Powers IntBot's Jose (San Jose airport) and Nylo (CES 2026) robots.

• Kodiak DriverFoundation model · Commercial

  Kodiak's SAE Level 4 autonomous-driving system for Class 8 trucks: sensing (LiDAR, radar, cameras in field-swappable SensorPods), redundant compute, and redundant steering/braking, running customer-owned trucks driverlessly. Compute on NVIDIA DRIVE; hardware scaled with Bosch.

• LeRobotFramework · Production · Open

  Hugging Face's open-source robot-learning framework. LeRobot is a framework, not a brain that powers robots directly: it HOSTS other models (pi0, pi0.5, GR00T N1.5) and provides datasets, simulation environments, training tooling, and a multi-GPU + plugin runtime. v0.4.0 added Datasets v3.0, LIBERO and Meta-World simulators, and the Robot Learning Course.

• Mentee BrainFoundation model · Research

  Mentee Robotics' end-to-end neural-network AI for the MenteeBot humanoid (Mentee Brain): integrates perception, reasoning, and motor control, trained via a proprietary Sim2Real pipeline, with NeRF-based on-the-fly environment mapping and transformer-based LLMs for interpreting commands and planning tasks. An AI-first stack spanning all operational layers. Mentee Robotics was acquired by Mobileye (Jan 2026) to lead its Physical AI division.

• Mi-Sense (Xiaomi)Research model · Research

  Xiaomi's perception and interaction AI for the CyberOne humanoid (unveiled 2022): the Mi-Sense depth-vision module for 3D spatial perception and the MiAI engine for vocal-emotion and environmental-sound recognition. Recognizes 45 human-emotion classifications and 85 environmental sounds, with real-time facial-expression and gesture detection. A research/showcase-stage system rather than a generalist control foundation model.

• Neuraverse Cognitive StackOS-layer · Pilot

  NEURA Robotics' hybrid neural plus symbolic multimodal cognitive stack. Neuraverse is a shared OS and learning platform connecting robots so that one robot's learned skill propagates to others. Positioned by NEURA as 'invisible OS for the World of Things.' At CES 2026, NEURA's robots are powered by NVIDIA Isaac GR00T XX, with Neuraverse acting as the orchestration layer atop GR00T.

• Nuro DriverFoundation model · Pilot

  Nuro's AI-driven Level 4 autonomy system (the Nuro Driver), built on NVIDIA DRIVE Thor, offered both as a full L4 stack and a Level 2++ Driver Assist. Validated across 8+ vehicle platforms with 1M+ autonomous miles and zero at-fault incidents; since Sept 2024 licensed to automotive OEMs and mobility providers alongside the Nuro AI Platform developer tools.

• OpenVLAResearch model · Research · Open

  An academic open-source vision-language-action model at 7B parameters. Multi-developer collaboration across Stanford, UC Berkeley, Toyota Research Institute, and Google DeepMind. Listed here as a research-model entity; architectural details should be firmed against the OpenVLA paper at build.

• pi0 (and pi0.5)Foundation model · Research · Open

  Physical Intelligence's open VLA flow model. A roughly 3B-parameter VLM backbone produces motor commands at up to 50 Hz across multiple robot platforms. Trained on internet-scale vision-language data, Open X-Embodiment, and Physical Intelligence's dexterous-manipulation dataset. The first robotics foundation model ported to Hugging Face LeRobot, open-sourced in February 2025 via the openpi repository under Apache-2.0. pi0 is the base of a fast-moving lineage: pi0-FAST (an autoregressive variant using the FAST action tokenizer), then pi0.5 (April 2025, open-world generalization, also released open in openpi), and the two newest and currently closed models, pi-star-0.6 (November 2025, which learns from experience via an RL method called Recap) and pi0.7 (April 2026, a roughly 5B-parameter steerable model on a Gemma3 backbone). Physical Intelligence raised a $600M Series B in November 2025 at a reported $5.6B valuation, led by Alphabet's CapitalG with Lux Capital and returning backers including Jeff Bezos, OpenAI, and Thrive Capital.

• pi0.5 (pi-0.5)Foundation model · Research · Open

  Physical Intelligence's pi0.5 (also written pi-0.5) is a vision-language-action model built on the company's pi0, adding open-world generalization by co-training on heterogeneous data including multi-robot trajectories, web data, high-level semantic subtask prediction, object detections, and verbal instructions, with a paper submitted on April 22, 2025. Its capability frontier is notable: it was the first end-to-end learned system demonstrated doing long-horizon dexterous manipulation, such as cleaning entire kitchens and bedrooms, in entirely new homes not present in its training data, running on mobile manipulators. It is recorded in the registry as built on pi0, the roughly 3-billion-parameter vision-language-action flow model that Physical Intelligence open-sourced in February 2025 via the openpi repository as the first robotics foundation model on Hugging Face's LeRobot. The registry records pi0.5 at research maturity: it was shown in research demonstrations in previously unseen homes rather than a named-customer rollout. Its open-source status is now confirmed: the pi05_base checkpoint is published in the openpi repository under Apache-2.0 (released after this record's initial ingest), so it is recorded as open, correcting the earlier closed-pending-confirmation note. pi0.5 sits between the open pi0 and the two newer closed models, pi-star-0.6 (November 2025) and pi0.7 (April 2026). This record also corrects the prior pi0 entry, which mis-dated pi0.5 to September 2025; the paper is April 22, 2025.

• Plus SuperDriveFoundation model · Pilot

  Plus's SAE Level 4 'virtual driver' (SuperDrive): a factory-installable autonomous-driving system designed for OEM integration into production trucks. In development and supervised fleet-trial stage (NVIDIA world-model collaboration; SuperDrive 6.0). No verified driver-out commercial freight yet; driverless validation targeted 2026.

• PonyWorld (Pony.ai Virtual Driver)World model · Commercial

  Pony.ai's proprietary world model (PonyWorld) and Virtual Driver technology, which together power its Gen-7 robotaxi and robotruck autonomous-driving stack. PonyWorld 2.0 (Apr 2026) is a self-improving physical-AI engine with self-diagnosis, targeted data collection in weak scenarios, and training focused on the hardest cases.

• Redwood AIFoundation model · Pilot

  Redwood AI is 1X Technologies' onboard vision-language-action control model for the NEO humanoid, recorded in the registry as a brain distinct from the separate 1X World Model. A roughly 160-million-parameter vision-language transformer tailored to the humanoid form factor, Redwood runs fully onboard NEO's embedded GPU at about 5 Hz and serves as the deployed real-time control and action policy, performing end-to-end mobile bi-manual manipulation, whole-body control, navigation, retrieving objects, and opening doors, trained on teleoperated and autonomous EVE and NEO episodes including learning from failure rollouts. It is distinct from the 1X World Model, a generative physics-grounded world model released in January 2026 for prediction, training, and evaluation: Redwood is the on-robot controller while the World Model is a generative simulation and learning substrate whose outputs are translated to motion by a separate inverse-dynamics model, so the two differ in model class, deployment role, release date, and training method rather than one being a rebrand of the other. Released around June 10, 2025, Redwood is closed and proprietary and is recorded at pilot maturity: it runs on real early-access NEO units offered at twenty thousand dollars or four hundred ninety-nine dollars per month, but 1X describes it as early in development that does not always succeed on the first try, and autonomous operation in customer homes is a stated goal rather than verified, so it is not presented as mass-shipped production.

• RFM-1Foundation model · Research

  Covariant (founded 2017 in the Berkeley area by Pieter Abbeel, Peter Chen, Rocky Duan, and Tianhao Zhang) built RFM-1, the Robotics Foundation Model announced in March 2024: a roughly 8-billion-parameter multimodal any-to-any sequence model trained on text, images, video, robot actions, and sensor readings, used to power warehouse pick-and-place robots. The registry records the brain at commercial maturity to reflect genuine deployed wiring: pre-acquisition, the Covariant Brain ran on real robotic arms at named customers including Otto Group, Radial, and McKesson, not just in demonstrations. Its corporate state is a key big-tech-absorption datapoint: in a reverse-acquihire that closed on August 30, 2024, Amazon hired founders Abbeel, Chen, and Duan along with about a quarter of staff and took a non-exclusive license to Covariant's models, reportedly for $380 million plus a $20 million payment per an FTC whistleblower complaint, with Covariant continuing as a diminished but independent company (NOT an Amazon subsidiary) under CEO Ted Stinson and remaining co-founder Tianhao Zhang, retaining its IP, while Amazon is a non-exclusive licensee rather than the brain developer; the Covariant Brain remains deployed via the KNAPP Pick-it-Easy OEM channel at customers such as McKesson, Wurth, GXO, and Brodrene Dahl. RFM-1 is proprietary with no peer-reviewed paper, model card, or released checkpoints, the human-like-reasoning framing is marketing, and current independent deployments and whether RFM-1 itself runs inside Amazon are not verified.

• RoboForce FMFoundation model · Pilot

  RoboForce's robot foundation model. Lighter public disclosure as of this pass. Entity created with a verified-vs-claimed depth note; firm primary sources at next pass.

• RobotEra ERA-42Foundation model · Pilot

  RobotEra's end-to-end Vision-Language-Action model, ERA-42 (introduced late 2024), described as the world's first truly embodied large model for five-fingered robot control. Integrates vision, language, and proprioceptive data in a unified on-device neural network for high-frequency, real-time motor control without cloud assistance; powers the RobotEra L7 humanoid. Developed under founder Chen Jianyu (Tsinghua University).

• RT-2 / RT-XResearch model · Research

  Google DeepMind's foundational 2023 research VLA. RT-2 is a transformer VLA trained on web text and images that directly outputs robot actions. Instantiations on PaLM-E and PaLI-X. RT-2-X is a 55B-parameter variant. Chain-of-thought reasoning for long-horizon planning. Superseded by Gemini Robotics for commercial paths, but conceptually ancestral to GR00T, Helix, pi0, and the broader dual-system descendants.

• Shield AI HivemindFoundation model · Production

  Shield AI's AI pilot and autonomy stack: it flies aircraft without GPS, communications, or a human pilot, and is offered as the Hivemind Enterprise developer platform (EdgeOS, Pilot, Commander, Forge). Has flown the X-62 VISTA (F-16-class) under DARPA ACE and the Anduril Fury, and powers the V-BAT.

• Skild BrainFoundation model · Commercial

  Skild AI's unified omni-bodied foundation model. Designed to control any robot without prior knowledge of body form (quadrupeds, humanoids, tabletop arms, mobile manipulators). Trained on online human videos and physics simulations across thousands of form factors. Built-in force-limiting safety constraints. Skild AI builds the robot brain, not hardware; platform strategy is to power other companies' robots.

• Skydio Autonomy EngineFoundation model · Production

  Skydio's onboard AI autonomy stack: vision-based obstacle avoidance, scene understanding, and self-flying skills that operate the drone independently of pilot stick input. The capability that distinguishes Skydio from pilot-assist drones.

• Stack AV Autonomous Driving SystemFoundation model · Research

  Stack AV's Level-4 autonomous-driving system for Class 8 trucks: a multi-sensor stack (lidar, radar, camera, AI-based perception). It has no separate product brand name and is development-stage; no verified driverless commercial operation as of 2026-05-30.

• Tesla FSD-BotFoundation model · Pilot

  Tesla's neural-net brain derived from the FSD vehicle stack plus Dojo training infrastructure. V3 is expected to run on Tesla's AI5 inference chip (taped out April 2026); Musk states AI5 has roughly 5x the memory bandwidth of its predecessor (AI4) and will ship first in Optimus, though these remain Musk-stated figures pending independent verification. On-device, vision-based, sharing architecture with Tesla vehicles. The Tesla AI-silicon advantage (FSD plus Dojo) is a real engineering edge. Less specified than Figure 03's stack was at the equivalent stage; full V3 specs are unknown as of mid-2026, with a summer 2026 unveil expected. Verified-vs-claimed: FSD-as-precedent is mixed; 'almost done' for nearly a decade.

• Torc Virtual DriverFoundation model · Pilot

  Torc's Level 4 autonomy software suite (the Virtual Driver): middleware/OS, perception, prediction, planning, motion control, and mission control. It runs on Daimler Truck's autonomous-ready Freightliner Cascadia; architecturally described as propulsion-agnostic, but it is integrated only on Freightliner today. Compute on NVIDIA (DRIVE). Pre-commercial: driver-out validated on a closed course (2024), supervised freight pilots ongoing, commercial launch target 2027.

• TuSimple Autonomous Driving SystemFoundation model · Research

  TuSimple's Level-4 autonomous-driving stack for long-haul trucks. Reached freight pilots and isolated driver-out demonstration runs but never sustained observer-free commercial. US self-driving operations were wound down after the company's 2024 delisting; the company rebranded to CreateAI and pivoted away from trucking.

• UBTech BrainNet (Thinker)Foundation model · Pilot

  UBTech's embodied-AI system for the Walker S2 industrial humanoid: the BrainNet 2.0 platform (multimodal reasoning, environmental understanding, and high-level task orchestration over stereo vision, force-feedback joints, and IMUs), combined with the Thinker embodied-AI foundation model, the Thinker-WM world model, and the Co-Agent multi-robot system, forming a dual-loop architecture for production-line swarm intelligence.

• UnifoLMWorld model · Research · Open

  Unitree Robotics' open-source unified-large-model series for general-purpose robot learning: comprising UnifoLM-WMA-0 (world-model-action) and UnifoLM-VLA-0 (vision-language-action). Unitree's G1 humanoid integrates UnifoLM-WMA-0.

• V-JEPA 2World model · Research · Open

  Meta's Fundamental AI Research lab released V-JEPA 2 on June 11, 2025, a self-supervised video world model of about 1.2 billion parameters trained on more than a million hours of video and a million images, in Yann LeCun's Joint-Embedding Predictive Architecture line. It is open, with code and checkpoints on GitHub and Hugging Face and an arXiv paper, and shipped alongside new physical-reasoning benchmarks. An action-conditioned variant, V-JEPA 2-AC, is post-trained on roughly 62 hours of unlabeled robot video from the open-source DROID dataset to enable zero-shot control. The registry records it at research maturity, the cleanest research world-model case in this expansion: it was demonstrated doing zero-shot pick-and-place of new objects in new environments at 65 to 80 percent success on robots in Meta's own labs, but there is no commercial robotics product, no named customers, and no robot-vendor wiring, so no registry robot should be wired to it under the verifiable-only-wiring rule since no real-robot product exists. Its strong open-artifact case of published weights, paper, and benchmarks clears the registrable bar easily, while the deployed-wiring caveat is explicit: this is demonstrated lab capability, not deployment, the lab success numbers are not shown to generalize beyond the tested arms, and the manipulation is short-horizon rather than long-horizon.

• VLT (XPENG VLA / VLA 2.0)Foundation model · Pilot · Open

  Xpeng Robotics' vision-centric VLA (Tesla-FSD-style, vision-only) shared across Xpeng EVs, robotaxis, and the IRON humanoid. Trained on a 30,000+ GPU cloud cluster. Runs on the Xpeng Turing AI chip. Announced for open-sourcing to global partners, with Volkswagen as the launch partner. VLA 2.0 rolled to Xpeng Ultra vehicles in Q1 2026; IRON humanoid mass production is targeted for end-2026.

• Waabi DriverFoundation model · Pilot

  Waabi's end-to-end, generative-AI L4 autonomy system for trucks: a single interpretable foundation-model-style stack trained and validated in Waabi World, a generative simulator (the 'teacher' that trains the Waabi Driver 'student'). Includes the Copilot4D world model. Integrated into the Volvo VNL Autonomous; runs with a safety driver pending driver-out validation.

• Waymo DriverFoundation model · Commercial

  Waymo's autonomous driving system (the Waymo Driver), now in its sixth generation, integrating a 360-degree sensor suite (cameras, lidar, radar) with perception, prediction, and planning to run fully-driverless robotaxi service. Waymo's research includes EMMA, a Gemini-powered end-to-end multimodal driving model that remains research-stage and is not deployed in production vehicles.

• Wayve AI Driver (and GAIA)Foundation model · Pilot

  Wayve (founded 2017 in London) is an embodied-AI and autonomous-driving company that raised a $1.05 billion Series C led by SoftBank in May 2024, the largest-ever UK AI round, followed by a roughly $1.2 billion Series D in February 2026 at a reported $8.6 billion post-money valuation, with total committed capital reaching about $1.5 billion once Uber's milestone-based commitment is included, backed by Eclipse, Balderton, and SoftBank Vision Fund 2 alongside strategic investors Nvidia, Uber, Microsoft, Mercedes-Benz, Nissan, and Stellantis. Its deployable brain is the Wayve AI Driver, an AV2.0 end-to-end embodied driving foundation model, complemented by the GAIA-1, GAIA-2, and GAIA-3 generative world models used as simulation, training, and evaluation tools, with GAIA-2 a latent-diffusion model documented in an arXiv report, and the LINGO vision-language-action layer for interpretability. The registry records it at pilot maturity with a load-bearing distinctness note: unlike the operator-captive robotaxi stacks already in the registry such as Waymo's and Nuro's drivers, Wayve is a foundation-model-first, embodiment- and OEM-agnostic licensed brain sold into multiple automakers rather than operating its own fleet, which is the registrable distinction. Its deployed wiring is genuinely on vehicles but at the piloting stage: a Nissan ProPilot integration is targeted for fiscal 2027, an Uber and Nissan robotaxi memorandum targets a Tokyo pilot in late 2026 on Nissan LEAFs running Nvidia DRIVE, a London robotaxi pilot is underway, and a Stellantis agreement signed in May 2026 brings the AI Driver into STLA AutoDrive with first North American vehicle integration targeted for 2028, while the GAIA world models themselves are training and simulation tools rather than on-vehicle policies. The OEM integrations are contracted or announced rather than in production, so the deployed wiring is piloting rather than at-scale commercial, and exact production-vehicle counts and whether any AI Driver deployment is currently driverless-commercial are not verified.

• WeRide OneFoundation model · Commercial

  WeRide's universal autonomous-driving platform (WeRide One), embedding an end-to-end AI model that spans Level 2 to Level 4 across all WeRide vehicle types (robotaxi, robobus, robovan, robosweeper). Designed to be decoupled from specific compute and chips for rapid migration across hardware, reducing OEM adaptation cost and deployment time.

• Wing OpenSkyOS-layer · Production

  Wing's airspace-management and automated-airspace-access platform (an FAA-approved UAS Service Supplier providing LAANC; CASA-verified in Australia). OpenSky coordinates drone traffic and airspace access for Wing's delivery network; Wing's onboard delivery-autonomy stack is not separately branded.

• Zipline Autonomy (Acoustic Detect-and-Avoid)Foundation model · Production

  Zipline's onboard autonomy stack and acoustic detect-and-avoid (DAA) system: a wing-mounted microphone array giving roughly 360-degree awareness up to about 2 km, plus ADS-B, enabling beyond-visual-line-of-sight flight without ground visual observers. Basis of Zipline's FAA BVLOS authorizations.

• Zoox Driving SystemFoundation model · Pilot

  Zoox's in-house AI driving system: the perception-prediction-planning stack at the center of its purpose-built, bidirectional robotaxi (no steering wheel or pedals). Predicts trajectories of vehicles, people, and animals up to about 8 seconds ahead with 360-degree sensing; powers Zoox's public driverless demonstration rides in Las Vegas and San Francisco.

Machine-readable: this page as markdown.