# AI stacks compared: what actually drives each robot

**Verdict:** The AI brain that drives a robot is a different thing from its body, and the stacks diverge sharply. Some are captive: Figure's Helix is a proprietary dual-system vision-language-action model that runs only Figure's own robots, and 1X's Redwood and Tesla's FSD-derived stack follow the same vertically integrated pattern. Others are open and licensable: NVIDIA's GR00T and Physical Intelligence's pi0 are foundation models other makers can build on. Many driver stacks (Waymo, Zoox, Nuro, WeRide) keep their architecture undisclosed, which DEPLOY records as an honest gap rather than guessing.

**Summary:** 59 verified brain records · 45 power at least one robot in the registry · 8 open source · 32 with architecture not disclosed.

Brain records are sourced from the DEPLOY registry. [How we verify](/methodology.md).

## Captive, third-party, or research-only

Beyond what kind of model a brain is, the sharpest axis is who can use it. DEPLOY's framework names four provider architectures:

- **Captive (vertically integrated):** Brain entity exists under the same corporate state as the hardware it powers. Not licensable to third parties; vertically integrated. The brain entity's roadmap, capabilities, and commercial trajectory are inseparable from the hardware roadmap.
- **Third-party foundation model:** Brain entity licensed or deployed across multiple unaffiliated hardware partners. Foundation-model-tier capability scope (cross-domain manipulation, cross-platform deployment). The brain entity's roadmap is partner-agnostic; commercial trajectory tied to broad adoption rather than a single hardware customer.
- **Hybrid (captive + third-party):** Brain entity has both captive-deployment AND third-party-licensing channels. The brain powers a parent's hardware AND is licensable to unaffiliated partners. Distinct from pure-captive (no licensing) and pure-third-party (no captive hardware).
- **Research-only:** Brain entity in development or laboratory stage; no verified commercial deployment. May target captive or third-party future architecture but the framework records the current state as research, not commercial.

## Foundation models and VLAs

Vision-language-action and end-to-end driving models: the generalist stacks that translate perception into motor action.

| Brain | Architecture | Access | Developer | Powers these robots | DEPLOY maturity |
|---|---|---|---|---|---|
| [Aurora Driver](/brains/aurora-driver.md) | Not disclosed | Proprietary | [Aurora Innovation](/companies/aurora.md) | [Aurora Driver on Peterbilt 579 (PACCAR)](/models/aurora-peterbilt-579.md), [Aurora Driver on International LT (2nd-gen)](/models/aurora-international-lt.md), [Volvo VNL Autonomous](/models/volvo-vnl-autonomous.md), [Aurora Driver](/models/aurora-driver.md) | Commercial |
| [Tesla FSD-Bot](/brains/tesla-fsd-bot.md) (Captive (vertically integrated)) | Neural-net brain derived from the Tesla FSD stack plus Dojo training. V3 runs on the Tesla AI5 chip (roughly 5x memory bandwidth of predecessor). On-device, vision-based, shares architecture with Tesla vehicles. | Proprietary | [Tesla](/companies/tesla.md) | [Tesla Optimus](/models/tesla-optimus.md), [Tesla Robotaxi (Model Y ADS)](/models/tesla-robotaxi.md), [Cybercab](/models/tesla-cybercab.md), [Semi](/models/tesla-semi.md) | Pilot |
| [Nuro Driver](/brains/nuro-driver.md) | Not disclosed | Proprietary | [Nuro](/companies/nuro.md) | [Nuro R2](/models/nuro-r2.md), [Nuro R3](/models/nuro-r3.md), [Nuro Autonomous Prius](/models/nuro-prius.md) | Pilot |
| [Skydio Autonomy Engine](/brains/skydio-autonomy-engine.md) | Not disclosed | Proprietary | [Skydio](/companies/skydio.md) | [Skydio X10](/models/skydio-x10.md), [Skydio X10D](/models/skydio-x10d.md), [Skydio Dock](/models/skydio-dock.md) | Production |
| [Avride ADS](/brains/avride-ads.md) | Not disclosed | Proprietary | [Avride](/companies/avride.md) | [Avride Robotaxi (Hyundai Ioniq 5)](/models/avride-robotaxi.md), [Avride delivery robot](/models/avride-delivery-robot.md) | Pilot |
| [Embark Driver](/brains/embark-driver.md) | Not disclosed | Proprietary | [Embark Trucks](/companies/embark-trucks.md) | [Embark Autonomous Truck](/models/embark-truck.md), [Embark Driver](/models/embark-driver.md) | Research |
| [Gatik Driver](/brains/gatik-driver.md) | Not disclosed | Proprietary | [Gatik](/companies/gatik.md) | [Gatik Autonomous Box Truck](/models/gatik-box-truck.md), [Gatik Autonomous Drive](/models/gatik-drive.md) | Commercial |
| [Genie Operator-1 (GO-1)](/brains/agibot-go-1.md) | Not disclosed | Proprietary | [AgiBot](/companies/agibot.md) | [AgiBot Yuanzheng A2](/models/agibot-yuanzheng-a2.md), [AgiBot Lingxi X2](/models/agibot-lingxi-x2.md) | Pilot |
| [GR00T N1 (Isaac GR00T)](/brains/gr00t-n1.md) | Dual-system VLA. System 2: VLM (NVIDIA-Eagle + SmolLM-1.7B) at roughly 10 Hz. System 1: diffusion transformer producing real-time motor actions. Jointly trained end-to-end on real-robot trajectories, human videos, and synthetic data. Runs on Jetson AGX Thor at the edge. | Open source | [NVIDIA](/companies/nvidia.md) | [AEON](/models/aeon.md), [Digit](/models/digit.md) | Research |
| [Helix (and Helix-02)](/brains/helix.md) | Dual-system VLA. System 2 (S2): onboard internet-pretrained VLM, 7B params, 7-9 Hz. System 1 (S1): fast reactive visuomotor policy, 80M params, 200 Hz. Helix-02 added System 0, a 1 kHz neural prior trained on 1,000+ hours of human motion data, replacing roughly 109,504 lines of hand-engineered C++. Runs on embedded low-power onboard GPUs, no cloud. | Proprietary | [Figure AI](/companies/figure-ai.md) | [Figure 02](/models/figure-02.md), [Figure 03](/models/figure-03.md) | Commercial |
| [Kodiak Driver](/brains/kodiak-driver.md) | Not disclosed | Proprietary | [Kodiak AI](/companies/kodiak.md) | [Kodiak RoboTruck](/models/kodiak-robotruck.md), [Kodiak Driver](/models/kodiak-driver.md) | Commercial |
| [Plus SuperDrive](/brains/plus-superdrive.md) | Not disclosed | Proprietary | [Plus](/companies/plus-ai.md) | [Plus SuperDrive Autonomous Truck](/models/plus-superdrive-truck.md), [PlusDrive](/models/plus-drive.md) | Pilot |
| [Shield AI Hivemind](/brains/hivemind.md) | Not disclosed | Proprietary | [Shield AI](/companies/shield-ai.md) | [Shield AI V-BAT](/models/shield-ai-v-bat.md), [Anduril Fury (YFQ-44A)](/models/anduril-fury.md) | Production |
| [UBTech BrainNet (Thinker)](/brains/ubtech-brainnet.md) | Not disclosed | Proprietary | [UBTech Robotics](/companies/ubtech-robotics.md) | [Walker S2](/models/ubtech-walker-s2.md), [Walker S Lite](/models/ubtech-walker-s-lite.md) | Pilot |
| [VLT (XPENG VLA / VLA 2.0)](/brains/vlt.md) | 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. | Open source | [XPeng Robotics](/companies/xpeng-robotics.md) | [XPeng IRON](/models/xpeng-iron.md), [XPENG Robotaxi](/models/xpeng-robotaxi.md) | Pilot |
| [Waabi Driver](/brains/waabi-driver.md) | Not disclosed | Proprietary | [Waabi](/companies/waabi.md) | [Volvo VNL Autonomous](/models/volvo-vnl-autonomous.md), [Waabi Driver Autonomous Truck](/models/waabi-truck.md) | Pilot |
| [Waymo Driver](/brains/waymo-driver.md) | Not disclosed | Proprietary | [Waymo](/companies/waymo.md) | [Waymo Driver (6th gen)](/models/waymo-driver-gen6.md), [Waymo Via (autonomous trucking)](/models/waymo-via.md) | Commercial |
| [Zipline Autonomy (Acoustic Detect-and-Avoid)](/brains/zipline-autonomy.md) | Not disclosed | Proprietary | [Zipline](/companies/zipline.md) | [Zipline Platform 1 (Zip)](/models/zipline-platform-1.md), [Zipline Platform 2 (P2)](/models/zipline-platform-2.md) | Production |
| [Apollo ADFM (Baidu)](/brains/baidu-apollo.md) | Not disclosed | Proprietary | [Baidu](/companies/baidu.md) | [Apollo RT6](/models/apollo-rt6.md) | Commercial |
| [Atlas-Gemini](/brains/atlas-gemini.md) | Hybrid stack: Boston Dynamics Atlas control software (including Orbit skill-sharing) integrated with Google DeepMind Gemini Robotics foundation models for higher-level reasoning and learning. | Proprietary | [Boston Dynamics](/companies/boston-dynamics.md), [Google DeepMind](/companies/google-deepmind.md) | [Atlas](/models/boston-dynamics-atlas.md) | Pilot |
| [Bot Auto Autonomous Driving System](/brains/bot-auto-autonomy.md) | Not disclosed | Proprietary | [Bot Auto](/companies/bot-auto.md) | [Bot Auto Autonomous Truck](/models/bot-auto-truck.md) | Commercial |
| [Carbon](/brains/carbon.md) | Cognitive hybrid translating natural language into precise physical actions. Emphasis on human-like hand dexterity for fine-manipulation tasks. | Proprietary | [Sanctuary AI](/companies/sanctuary-ai.md) | [Phoenix](/models/sanctuary-phoenix.md) | Commercial |
| [Einride Driver](/brains/einride-driver.md) | Not disclosed | Proprietary | [Einride](/companies/einride.md) | [Einride Pod](/models/einride-pod.md) | Commercial |
| [Gemini Robotics](/brains/gemini-robotics.md) | VLA built on Gemini 2.0 with physical actions as output modality, plus an intermediate reasoning layer for spatial analysis and safety. Gemini Robotics-ER is the embodied-reasoning VLM companion. Gemini Robotics On-Device runs locally, network-independent, and is fine-tunable with 50-100 demonstrations. Gemini Robotics 1.5 emphasizes a 'think before acting' chain-of-thought style. | Proprietary | [Google DeepMind](/companies/google-deepmind.md) | [Apollo](/models/apptronik-apollo.md) | Research |
| [GraspVLA (Galbot)](/brains/graspvla.md) | Not disclosed | Proprietary | [Galbot](/companies/galbot.md) | [Galbot G1](/models/galbot-g1.md) | Research |
| [Grok (xAI)](/brains/grok.md) | Large language model (System 2 reasoning) deployed as the conversational and high-level instruction layer of Tesla Optimus's dual-brain stack. Paired with Tesla's FSD-derived visuomotor neural networks (System 1, see /brains/tesla-fsd-bot). | Proprietary | [xAI](/companies/xai.md) | [Tesla Optimus](/models/tesla-optimus.md) | Pilot |
| [IntEngine (IntBot)](/brains/intengine.md) | Not disclosed | Proprietary | [IntBot](/companies/intbot.md) | [José (IntBot humanoid)](/models/intbot-jose.md) | Pilot |
| [Mentee Brain](/brains/mentee-brain.md) | Not disclosed | Proprietary | [Mentee Robotics](/companies/mentee-robotics.md) | [MenteeBot](/models/menteebot.md) | Research |
| [Redwood AI](/brains/1x-redwood.md) (Captive (vertically integrated)) | 160M-parameter vision-language-action (VLA) transformer; onboard, ~5Hz; end-to-end mobile manipulation | Proprietary | [1X Technologies](/companies/1x-technologies.md) | [NEO](/models/1x-neo.md) | Pilot |
| [RobotEra ERA-42](/brains/robotera-era.md) | Not disclosed | Proprietary | [Robot Era](/companies/robot-era.md) | [RobotEra L7](/models/robotera-l7.md) | Pilot |
| [Stack AV Autonomous Driving System](/brains/stack-av-autonomy.md) | Not disclosed | Proprietary | [Stack AV](/companies/stack-av.md) | [Stack AV Autonomous Truck (Peterbilt 579)](/models/stack-av-truck.md) | Research |
| [Torc Virtual Driver](/brains/torc-virtual-driver.md) | Not disclosed | Proprietary | [Torc Robotics](/companies/torc-robotics.md) | [Autonomous Freightliner Cascadia (Torc)](/models/torc-freightliner-cascadia.md) | Pilot |
| [TuSimple Autonomous Driving System](/brains/tusimple-driver.md) | Not disclosed | Proprietary | [TuSimple](/companies/tusimple.md) | [TuSimple Autonomous Truck](/models/tusimple-truck.md) | Research |
| [WeRide One](/brains/weride-one.md) | Not disclosed | Proprietary | [WeRide](/companies/weride.md) | [Robotaxi GXR](/models/weride-gxr.md) | Commercial |
| [Zoox Driving System](/brains/zoox-driver.md) | Not disclosed | Proprietary | [Zoox](/companies/zoox.md) | [Zoox Robotaxi](/models/zoox-robotaxi.md) | Pilot |
| [DYNA-1](/brains/dyna-1.md) | Dexterous robot foundation model (VLA) for sustained autonomous operation on a pair of stationary dual robotic arms (the Dyna Robotics 'Dynasaur' system, not humanoid). | Proprietary | [Dyna Robotics](/companies/dyna-robotics.md) | No registry robot yet | Commercial |
| [Field Foundation Models (FFMs)](/brains/field-foundation-models.md) | Physics-first, embodiment-agnostic field/mobility autonomy foundation models for GPS-/map-denied unstructured environments (navigation + mobile autonomy without maps/GPS/pre-defined trajectories). | Proprietary | [Field AI](/companies/field-ai.md) | No registry robot yet | Commercial |
| [GEN-0 (and GEN-1)](/brains/gen-0.md) | Embodied/manipulation foundation model (generalist VLA-class) trained on raw physical-interaction data; ~7B-param scaling phase transition reported. | Proprietary | [Generalist AI](/companies/generalist-ai.md) | No registry robot yet | Research |
| [pi0 (and pi0.5)](/brains/pi0.md) | VLA flow model, roughly 3B-parameter VLM backbone, motor commands at up to 50 Hz, cross-embodiment across 8 platforms: UR5e, Bimanual UR5e, Franka, Bimanual Trossen, Bimanual ARX, Mobile Trossen, Mobile Fibocom. Trained on internet-scale vision-language data, Open X-Embodiment, and Physical Intelligence's dexterous-manipulation dataset. PyTorch and JAX implementations. | Open source | [Physical Intelligence](/companies/physical-intelligence.md) | No registry robot yet | Research |
| [pi0.5 (pi-0.5)](/brains/pi05.md) | Vision-language-action model built on pi0; co-trained on heterogeneous multi-robot + web data with high-level semantic subtask prediction for open-world generalization. | Open source | [Physical Intelligence](/companies/physical-intelligence.md) | No registry robot yet | Research |
| [RFM-1](/brains/rfm-1.md) | Multimodal any-to-any autoregressive sequence model (~8B params); next-token prediction over text/images/video/robot-actions/sensors. | Proprietary | [Covariant](/companies/covariant.md) | No registry robot yet | Research |
| [RoboForce FM](/brains/roboforce-fm.md) | Robot foundation model; specific architectural details are limited in public disclosure this pass and should be firmed at next pass. | Proprietary | [RoboForce](/companies/roboforce.md) | No registry robot yet | Pilot |
| [Robostral Navigate](/brains/robostral-navigate.md) | 8B parameter VLM initialized from vision-language model for grounding (pointing, counting, object localization). Navigation via pointing (infers image coordinates of target location) with fallback to metric displacements. Trained with prefix-caching (tree-based attention masking, 22x token reduction) and online RL (CISPO algorithm, +3.2% improvement). | Proprietary | [Mistral AI](/companies/mistral-ai.md) | No registry robot yet | Research |
| [Skild Brain](/brains/skild-brain.md) | Unified 'omni-bodied' foundation model controlling any robot form without prior body-form knowledge (quadrupeds, humanoids, tabletop arms, mobile manipulators). Trained on online human videos plus physics simulations across thousands of form factors. Built-in force-limiting safety constraints. | Proprietary | [Skild AI](/companies/skild-ai.md) | No registry robot yet | Commercial |
| [Wayve AI Driver (and GAIA)](/brains/wayve-ai-driver.md) | End-to-end embodied driving foundation model (AV2.0); paired with GAIA latent-diffusion generative world models (sim/training) + LINGO VLA interpretability layer. | Proprietary | [Wayve](/companies/wayve.md) | No registry robot yet | Pilot |

## World models

Models that predict how the physical world evolves, used for planning, simulation, and policy learning.

| Brain | Architecture | Access | Developer | Powers these robots | DEPLOY maturity |
|---|---|---|---|---|---|
| [UnifoLM](/brains/unifolm.md) | Unitree's open-source unified-large-model series for general-purpose robot learning across multiple embodiments. UnifoLM-WMA-0 (world-model–action): a world model of robot–environment physical interaction, used both as a simulation engine for synthetic data and as a policy-enhancement head that predicts future environmental states (video prediction translated into limb actions). UnifoLM-VLA-0 (vision–language–action): a VLA for general-purpose humanoid manipulation, built on a UnifoLM-VLM base. Open-sourced on Unitree's GitHub (WMA-0 Sept 2025; VLA-0 Jan 2026). | Open source | [Unitree Robotics](/companies/unitree-robotics.md) | [Unitree G1](/models/unitree-g1.md), [Unitree R1](/models/unitree-r1.md), [Unitree H1](/models/unitree-h1.md), [Unitree H2](/models/unitree-h2.md) | Research |
| [1X World Model](/brains/1x-world-model.md) | Generative world model enabling learning by predicting outcomes versus exhaustive pre-programming; combined with VLA control for action generation. | Proprietary | [1X Technologies](/companies/1x-technologies.md) | [NEO](/models/1x-neo.md), [EVE](/models/1x-eve.md) | Commercial |
| [PonyWorld (Pony.ai Virtual Driver)](/brains/pony-world.md) | Not disclosed | Proprietary | [Pony AI](/companies/pony-ai.md) | [Pony Gen-7 Robotaxi](/models/pony-gen7-robotaxi.md) | Commercial |
| [V-JEPA 2](/brains/vjepa-2.md) | Self-supervised joint-embedding predictive video world model (~1.2B params); action-conditioned V-JEPA 2-AC variant for robot control. | Open source | [Meta](/companies/meta.md) | No registry robot yet | Research |

## Frameworks

Open robot-learning frameworks that host and train other brains rather than drive a robot directly.

| Brain | Architecture | Access | Developer | Powers these robots | DEPLOY maturity |
|---|---|---|---|---|---|
| [AeroVironment AVACORE](/brains/avacore.md) | Not disclosed | Proprietary | [AeroVironment](/companies/aerovironment.md) | [AeroVironment Puma](/models/aerovironment-puma.md), [AeroVironment P550](/models/aerovironment-p550.md) | Production |
| [LeRobot](/brains/lerobot.md) | Open-source robot-learning framework. Library-style hosting of VLA models (pi0, pi0.5, GR00T N1.5) plus datasets, simulation environments (LIBERO, Meta-World), training tooling, multi-GPU support, plugin system. Released v0.4.0 with the upgrades above. | Open source | [Hugging Face](/companies/hugging-face.md) | No registry robot yet | Production |

## OS-layers and autonomy stacks

Autonomy operating systems licensed to hardware makers as the driving layer of their machines.

| Brain | Architecture | Access | Developer | Powers these robots | DEPLOY maturity |
|---|---|---|---|---|---|
| [Anduril Lattice](/brains/lattice.md) | Not disclosed | Proprietary | [Anduril Industries](/companies/anduril.md) | [Anduril Roadrunner](/models/anduril-roadrunner.md), [Anduril Bolt](/models/anduril-bolt.md), [Anduril Fury (YFQ-44A)](/models/anduril-fury.md), [Ghost (Ghost-X)](/models/anduril-ghost.md), [Dive-LD](/models/anduril-dive-ld.md), [Ghost Shark (XL-AUV)](/models/anduril-ghost-shark.md) | Production |
| [BrainOS](/brains/brainos.md) | Cloud-connected autonomy OS: teach-and-repeat / mapped autonomous routing with onboard sensor-based obstacle avoidance, plus a cloud portal and mobile app for fleet management and reporting. Licensed to OEMs as the autonomy layer of their cleaning hardware. | Proprietary | [Brain Corp](/companies/brain-corp.md) | [SoftBank Whiz](/models/softbank-whiz.md), [Tennant X4 ROVR](/models/tennant-x4-rovr.md), [ICE Cobi 18](/models/ice-cobi-18.md), [Nilfisk Liberty SC60](/models/nilfisk-liberty-sc60.md), [Tennant T7AMR](/models/tennant-t7amr.md) | Production |
| [Mobileye Drive](/brains/mobileye-drive.md) | Not disclosed | Proprietary | [Mobileye](/companies/mobileye.md) | [Volvo EX90](/models/volvo-ex90.md), [Mobileye Drive](/models/mobileye-drive.md), [VW ID. Buzz AD](/models/vw-id-buzz-ad.md) | Pilot |
| [Neuraverse Cognitive Stack](/brains/neuraverse.md) | Hybrid neural and symbolic multimodal cognitive stack. Plus Neuraverse, a shared OS and learning platform that connects robots so that one robot's learned skill propagates to others. NEURA's CES 2026 robots are powered by NVIDIA Isaac GR00T XX; Neuraverse is the orchestration layer atop GR00T. | Proprietary | [NEURA Robotics](/companies/neura-robotics.md) | [4NE-1](/models/neura-4ne-1.md), [MiPA](/models/neura-mipa.md) | Pilot |
| [Wing OpenSky](/brains/wing-opensky.md) | Not disclosed | Proprietary | [Wing](/companies/wing.md) | [Wing Delivery Aircraft](/models/wing-delivery-aircraft.md) | Production |

## Research models

Research-stage brains not yet carried into a commercial product in the registry.

| Brain | Architecture | Access | Developer | Powers these robots | DEPLOY maturity |
|---|---|---|---|---|---|
| [Mi-Sense (Xiaomi)](/brains/mi-sense.md) | Not disclosed | Proprietary | [Xiaomi](/companies/xiaomi.md) | [Xiaomi CyberOne](/models/xiaomi-cyberone.md) | Research |
| [OpenVLA](/brains/openvla.md) | 7B-parameter VLA. Architecture and training-data details pending firmer primary-source confirmation (OpenVLA paper at openvla.github.io and arXiv). | Open source | [Google DeepMind](/companies/google-deepmind.md) | No registry robot yet | Research |
| [RT-2 / RT-X](/brains/rt-2.md) | Transformer VLA trained on web text and images, directly outputs robot actions. Instantiations on PaLM-E and PaLI-X. RT-2-X is 55B parameters. Chain-of-thought reasoning for long-horizon planning. Trained on web data plus Open X-Embodiment. | Proprietary | [Google DeepMind](/companies/google-deepmind.md) | No registry robot yet | Research |

## See also

- [Robot brains](/brains.md): every AI brain in the registry.
- [Humanoid robot autonomy: verified vs claimed](/humanoid-robot-autonomy-verified.md): claimed autonomy set against the verified record.
- [All humanoid robots](/humanoids.md): every humanoid in the registry.

_DEPLOY does not build robot brains. We verify what each stack is, who develops it, the robots it powers, and its maturity. Where a maker has not disclosed an architecture, we say so rather than guess._

_API: GET /v1/entities?type=brain · canonical URL: /ai-stacks_