Skip to content
Saved
Unitree Go2
Robot Tracker

Unitree Go2

Unitree · Specialty Robots
IIAssistedReady Now
  1. 2026
  2. Security researchers disclose Go2 firmware remote access vulnerabilities

    Security research published in 2024 documented vulnerabilities in Unitree Go2's firmware that allowed unauthenticated command execution over the robot's local Wi-Fi network.

    Full assessment
    AutonomyL2 unchanged
    ReadinessPromising progress strengthened
    ScoreReliability -5

    The research demonstrated that an attacker within network range could issue movement commands without authorization. Unitree's patch response timeline remained unclear; buyers operating Go2 in shared network environments should treat this as an active risk until a confirmed firmware update with documented remediation ships.

    Impact on autonomy

    • Vulnerability enables unauthorized movement commands bypassing operator intent
    • No autonomous safety stop documented for unauthorized command injection scenarios
    • Level II classification maintained: vulnerability affects control plane, not autonomy architecture
    • Operator cannot verify command provenance at the gait execution layer

    Impact on readiness

    • Shared-network deployments carry documented unauthorized control risk
    • No confirmed patch release with public changelog at time of disclosure
    • Field operators and research institutions advised to isolate Go2 on dedicated networks
    • Buyers in enterprise or multi-user environments face higher exposure than single-user home labs

    Claim check4 claims reviewed

    Developer-friendly open platform
    Open API architecture that enables developer access also expanded the attack surface; unauthenticated command paths documented by researchers
    Secure Wi-Fi operation for field deployments
    Researchers demonstrated unauthenticated command execution over Wi-Fi; secure operation requires network isolation not documented in standard setup guidance
    Manufacturer response to security findings
    Unitree acknowledged the disclosure; a fully patched firmware release with public changelog confirming remediation was not documented within the initial disclosure window
    Suitable for research institution deployment
    Institutional IT security review requirements raised by disclosure; network isolation is a prerequisite for compliant deployment per standard institutional policies

    Bottom lineThe disclosure is a setback for Go2's enterprise and institutional readiness; hobbyist single-network deployments carry lower exposure but the patch gap remains an open question.

    Technical notes4 sections
    Vulnerability Scope

    Published research documented unauthenticated remote command execution paths accessible via Go2’s Wi-Fi interface. The vulnerabilities allowed an attacker on the same network segment to send movement commands to the robot without valid credentials. The research covered the Go2 firmware version available at time of disclosure; specific version numbers were referenced in the original research but are not reproduced here to avoid fabrication of version strings.

    Attack Surface

    Go2’s developer-accessible Wi-Fi API, designed to enable SDK connectivity, was identified as the primary attack surface. The open architecture that makes Go2 valuable for research and custom integration also means authentication controls on that interface are critical. Researchers demonstrated movement command injection; potential for sensor data interception was noted as an additional risk vector.

    Remediation Status at Disclosure

    Unitree acknowledged the security findings. A patched firmware with documented remediation of the specific vulnerabilities was not confirmed as shipped within the reporting window following the disclosure. Operators should verify current firmware release notes against the disclosed vulnerability identifiers before field deployment.

    Operational Guidance

    Network isolation on a dedicated SSID or VLAN, not shared with other devices or users, reduces exposure. Remote operation over public or semi-public networks should be suspended pending confirmed patch deployment.

    SourcesCybereason Security Research Blog 2024-04-29The Register 2024-05-02
    Reliability
  3. 2024
  4. Unitree Go2-W wheeled variant announced, expanding platform lineup

    Unitree announced the Go2-W, a wheeled derivative sharing the Go2 body and compute stack but substituting a wheeled base for the quadruped leg assembly.

    Full assessment
    AutonomyL2 unchanged
    ReadinessPromising progress strengthened
    ScoreScores unchanged

    The announcement documents a platform expansion, not a capability advance for the Go2 itself. The Go2-W is a distinct product; its indoor mobility trade-off against the Go2's rough-terrain capability reflects different deployment contexts rather than iterative improvement.

    Impact on autonomy

    • Go2 (legged) autonomy level and architecture unchanged by this announcement
    • Go2-W uses wheeled locomotion; rough-terrain capability of the legged Go2 is not replicated
    • Shared compute stack between variants enables software portability for developers
    • Level II classification applies to the Go2; Go2-W classification is a separate assessment

    Impact on readiness

    • Go2 readiness status unchanged; announcement concerns a separate product
    • Developer investment in Go2 SDK transferable to Go2-W given shared architecture
    • Go2-W targets different deployment surface than the legged Go2; not a substitute
    • Availability timeline and pricing for Go2-W not confirmed at announcement

    Claim check4 claims reviewed

    Go2 platform expanded for more use cases
    Accurate in the sense of a variant product; the Go2 (legged) itself is unchanged. The Go2-W is a different locomotion architecture, not an enhancement of the original
    Wheeled variant improves indoor performance
    Wheels reduce complexity and cost for smooth-surface operation; the tradeoff is loss of the legged Go2's rough-terrain and stair navigation capability, however limited
    Same powerful compute platform
    Shared body and compute stack documented; locomotion controller software differs substantially between legged and wheeled configurations
    More accessible price point expected
    Pricing not confirmed at announcement; wheeled locomotion typically reduces component cost but Unitree did not publish Go2-W pricing at the announcement stage

    Bottom lineThe Go2-W announcement signals platform versatility strategy from Unitree; it does not document progress in the Go2 quadruped's own capability or commercial trajectory.

    Technical notes4 sections
    Go2-W Platform Relationship

    The Go2-W retains the Go2’s upper body structure, payload rail, and compute architecture while replacing the four-leg assembly with a wheeled base. This approach reduces the mechanical complexity of locomotion and eliminates the gait control demands of legged navigation, trading outdoor terrain capability for smoother and more predictable indoor mobility.

    Locomotion Differences

    The Go2 (legged) navigates rough terrain, slopes, and unstructured outdoor environments via its proprietary gait stack. The Go2-W operates on smooth and semi-smooth surfaces with conventional wheeled navigation. Stair climbing, which the legged Go2 handles with difficulty, is not a design target for the wheeled variant. Obstacle traversal capability of the wheeled platform is lower than the legged version on uneven terrain.

    Developer Implications

    Software developed for Go2 using ROS2 and Unitree’s SDK has partial portability to the Go2-W due to shared compute hardware. Locomotion-specific code must be rewritten for the wheeled controller. Sensor and payload integrations built on the shared payload rail architecture transfer without modification.

    Classification Note

    This entry documents the announcement as it relates to the Go2 product history. The Go2-W is a separate product requiring its own Robovations classification assessment.

    SourcesUnitree Robotics Official Announcement 2024-08-28IEEE Spectrum 2024-09-03
  5. 2023
  6. Unitree Go2 GPT integration enables voice-commanded locomotion demos

    Unitree published demonstration footage of Go2 executing locomotion commands issued via natural language, routed through a GPT API integration layer.

    Full assessment
    AutonomyL2 unchanged
    ReadinessPromising progress strengthened
    ScoreScores unchanged

    The demo showed the robot responding to verbal instructions with corresponding gait changes on flat indoor terrain. The integration relies on external cloud inference; latency and offline reliability are not documented for field use.

    Impact on autonomy

    • Natural language to gait translation demonstrated; requires cloud API connectivity
    • Command interpretation adds software layer above existing gait controller
    • Level II classification unchanged: a human issues every command, cloud or otherwise
    • No onboard language model; demonstrated capability is cloud-dependent

    Impact on readiness

    • Feature not released as stable SDK component; developer experimentation only
    • Cloud dependency limits use cases requiring offline or low-latency operation
    • Community implementation examples shared on GitHub; not officially supported

    Claim check4 claims reviewed

    Go2 understands and responds to human speech
    Demonstrated via external GPT API call; robot itself has no onboard language processing per available documentation
    Natural language commands enable intuitive control
    Demo footage shows controlled indoor environment; real-time latency and robustness in outdoor or noisy conditions not documented
    AI-powered locomotion control
    AI component is the cloud language model, not an onboard system; gait execution remains the existing scripted controller
    Significant advance in robot-human interaction
    Lateral addition: existing locomotion capability accessed through a new input modality; underlying autonomy level unchanged

    Bottom lineThe integration demonstrates how developer-accessible APIs can layer new interaction modes onto Go2; it does not document a capability advance in autonomous operation.

    Technical notes4 sections
    Integration Architecture

    Demonstration used Go2’s Wi-Fi connectivity and onboard compute to relay speech input to an external GPT API endpoint. The language model parsed the command and returned a structured instruction that the Go2 SDK translated into gait parameters. The full pipeline introduced cloud round-trip latency between spoken command and movement execution.

    Locomotion Commands Demonstrated

    Published footage included directional movement, sit and stand transitions, and gait speed changes issued via spoken phrases. All demonstrated behaviors were already within Go2’s existing SDK-accessible gait library; the novelty was the input modality, not the underlying motion capability.

    Autonomy Classification Impact

    The integration does not alter Go2’s Level II classification. Every movement command originates from a human operator; the natural language layer is a translation step, not an autonomous decision-making system. Level II: Assisted Autonomy, confirmed. The dependency on cloud inference latency also introduces a failure mode not present in direct SDK control.

    Developer Availability

    Third-party community implementations circulated on GitHub following the demonstration. Unitree did not ship this as an official SDK feature in the period following the demo. Developers replicating the integration reported variable latency depending on network conditions and API response time.

    SourcesUnitree Robotics YouTube Channel 2023-11-14The Robot Report 2023-11-22
  7. Unitree Go2 launches commercially at $1,600 starting price

    Unitree began fulfilling Go2 orders in July 2023 at a $1,600 starting price, making legged quadruped hardware commercially available outside laboratory procurement channels for the first time in this price range.

    Full assessment
    AutonomyL2 confirmed
    ReadinessPromising progress strengthened
    ScoreCommercial availability established

    The Air, Pro, and Edu tiers each ship with varying compute and sensor configurations.

    WatchingFor shipping lead times and warranty support coverage outside China.

    Impact on autonomy

    • Preprogrammed gait routines ship ready; custom behavior requires SDK development
    • Onboard sensors and API documented for third-party computer vision integration
    • Level II classification confirmed at launch: locomotion scripts, no end-to-end task autonomy

    Impact on readiness

    • Direct sales channel established; no third-party retail distribution at launch
    • Shipping from China documented at 2-4 weeks; no North American service center
    • Developer-oriented setup; not plug-and-play for non-technical buyers

    Claim check4 claims reviewed

    Consumer-ready quadruped starting at $1,600
    Priced accessibly for hobbyists but requires developer skills for meaningful operation; not consumer-grade plug-and-play per community reports
    Capable of autonomous outdoor navigation
    Out-of-the-box autonomy limited to preprogrammed gaits; outdoor mission autonomy requires custom SDK implementation
    Multi-tier options for all use cases
    Air, Pro, and Edu tiers differ in compute and sensors; Air tier lacks the onboard Jetson compute present in Pro and Edu configurations
    Rough terrain locomotion without operator input
    Terrain adaptation is gait-scripted; dynamic obstacle avoidance requires developer-coded behavior per Unitree SDK documentation

    Bottom lineGo2's launch price expanded access to legged robotics research hardware; the developer dependency narrows the practical audience compared to marketing scope.

    Technical notes4 sections
    Tier Configuration at Launch

    Go2 launched with three configurations: Air (base locomotion, no Jetson compute), Pro (onboard Jetson compute, upgraded sensor suite), and Edu (research-focused, full SDK and ROS2 support). Price range $1,600 to approximately $2,500 depending on tier and configuration.

    Locomotion Architecture

    Go2 uses Unitree’s proprietary gait control stack with preprogrammed locomotion routines for trot, bound, and crawl gaits. Terrain adaptation relies on IMU and leg-mounted sensors feeding the gait controller. The SDK exposes APIs for custom gait parameterization and external sensor integration via ROS2.

    Physical Specifications

    Dimensions at launch: approximately 27.6 in length, 11.2 in width, 15 in standing height. Weight reported at 26.5 lbs. Battery: 144 Wh Li-ion, rated approximately 40-60 minutes runtime depending on terrain and payload. Modular payload rail on top deck supports custom sensor and compute attachments up to documented payload limits.

    Connectivity

    Wi-Fi and Bluetooth. Remote control available via app; SDK connectivity over Wi-Fi for development workflows. No cellular module in base configuration.

    SourcesUnitree Robotics Official Store 2023-07-01IEEE Spectrum 2023-07-18