ROS2 Support in Humanoid Robots: A Developer's Guide (2026)
Which humanoid robots actually support ROS2? We tested SDK access, URDF availability, and MoveIt2 compatibility across all 43 robots in our database.
The ROS2 Landscape
ROS2 (Robot Operating System 2) has become the de facto middleware for robotics research and increasingly for production. But "ROS2 support" means different things to different manufacturers.
Tier 1: Native ROS2 (Full SDK + URDF + MoveIt2)
These robots ship with official ROS2 packages, URDF/MJCF models, and MoveIt2 configs:
- PAL TALOS β The gold standard. Full ROS2 Humble + Jazzy, official Gazebo sim, MoveIt2, Nav2. Years of validation.
- Pollen Reachy2 β Fully open-source ROS2 stack. Hugging Face backing. Best-in-class for researcher access.
- Unitree G1 / H1 β SDK2 with native ROS2 bridge. Official MuJoCo/Gazebo models. Largest community after TALOS.
- Booster T1 β ROS2 + Python/C++ SDKs. 50+ research teams using it via RoboCup.
Tier 2: Partial ROS2 (Compatible but limited)
- Agility Digit β ROS2 compatible but not the primary interface. Python SDK preferred.
- Fourier GR-1 / GR-2 β ROS + Isaac Lab. Growing community but fewer official tools.
- AgiBot A2 β AimRT framework bridges to ROS2/gRPC/MQTT. Chinese docs primarily.
Tier 3: No ROS2
- Tesla Optimus β Fully proprietary FSD-derived stack. Zero public API.
- Figure 02 / 03 β Proprietary Helix system. OpenAI API integration but no robotics middleware.
- 1X NEO β World Model AI, no public SDK.
- Boston Dynamics Atlas β NVIDIA Isaac internally but almost nothing public.
What to Look For
When evaluating ROS2 support, check:
1. URDF/MJCF availability β Without a robot model, simulation is guesswork.
2. MoveIt2 config β For manipulation tasks, this is essential.
3. Isaac Sim support β NVIDIA's sim is becoming the industry standard for RL training.
4. Community size β More users = more bug reports = better software.
Check our Integration Matrix for detailed compatibility ratings across ROS2, Isaac Sim, MuJoCo, Gazebo, and more.
ROS2 Distro Coverage β Humble vs Jazzy
The current ROS2 LTS landscape has two relevant releases:
- Humble (Hawksbill) β May 2022, LTS through 2027. The production default for robots shipping in 2024-2026.
- Jazzy (Jalisco) β May 2024, LTS through 2029. The forward-looking target. Required for newer Nav2 features and improved real-time scheduling.
Per-robot distro support among Tier 1 vendors:
| Robot | Humble | Jazzy | Rolling | Notes | |---|---|---|---|---| | PAL TALOS | β | β | dev | Cross-validated through PAL's own CI | | Pollen Reachy 2 | β | β | community | Hugging Face mirrors include both | | Unitree G1/H1 | β | partial | β | SDK2 ships Humble-first; Jazzy bindings in beta | | Booster T1 | β | community | β | Official Humble; community Jazzy fork | | Astribot S1 | β | β | β | Humble only, vendor-confirmed | | RAISE A1 | β | β | β | Humble only |If your team standardises on Jazzy for 2026+ deployments, Tier 1 narrows to PAL and Pollen. Most other "ROS2 native" robots are still Humble-only until Q4 2026 or later.
URDF and Simulation Models β The Make-or-Break Asset
A URDF (or modern MJCF) file is the single most valuable artifact a humanoid manufacturer can publish. Without it, simulating before purchase is impossible.
Of the 12 robots with native ROS2 support, here is URDF/MJCF availability:
- Public URDF on GitHub or vendor site (8): PAL TALOS, Pollen Reachy 2, Unitree G1/H1/H1-2/H2/R1, Booster T1
- URDF on NDA-restricted developer portal (3): Atlas Electric (limited partner program), Digit (Amazon-tier customers), AgiBot A2 (enterprise tier)
- MJCF only (no URDF): Booster T1 (also has URDF), some Unitree platforms (MJCF primary, URDF generated)
The "URDF on GitHub" group is the practical research-and-evaluation set. The NDA group is gated to procurement-stage customers, which makes pre-purchase evaluation a chicken-and-egg problem.
Recommendation for buyers: if you cannot get a URDF before signing a PO, you are buying blind. Walk away or demand a 14-day eval that includes the model files.
MoveIt2 + Nav2 β The Manipulation and Navigation Stack
ROS2 is the middleware. MoveIt2 and Nav2 are the stacks that actually let you do things with the robot.
MoveIt2 (manipulation) coverage:- β Working out-of-box: PAL TALOS, Pollen Reachy 2
- β Working with vendor config: Unitree G1/H1 (community MoveIt2 configs maintained by ETH Zurich and Open Robotics members), Booster T1
- β Possible with custom SRDF: Atlas Electric, Digit
- β Not officially supported: Tesla, Figure, 1X, most Chinese closed platforms
- β Vendor-shipped configs: TALOS, Digit (best-in-class with the indoor warehouse stack)
- β Community configs: Unitree, Booster
- β Partial: Reachy 2 (table-top, not mobile)
- β Not applicable / proprietary: Tesla, Figure, 1X
A practical takeaway: if your application is "humanoid that picks things in a warehouse and navigates aisles," your shortlist is Digit, TALOS, and Unitree H1/H2. Everything else either lacks the stack or hides it behind an enterprise sales cycle.
Isaac Sim Support β The RL Training Path
NVIDIA Isaac Sim has become the de facto target for reinforcement-learning policy training. Sim2real for humanoids requires accurate physics, accurate sensor models, and accurate actuator models.
Isaac Sim/Isaac Lab support by robot:
- First-class assets shipping in Isaac Lab repo: Unitree H1, Unitree G1, Boston Dynamics Spot (not a humanoid but reference), Pollen Reachy 2 (community contribution).
- Vendor-provided USD assets: Booster T1, Apptronik Apollo (partial β NVIDIA partnership), Figure 02 (Helix demo).
- Community USD assets: Several Chinese humanoids reverse-engineered from publicly available URDFs.
The Unitree H1 has become the de facto "Hello World" humanoid for RL research β it shows up in three out of four humanoid locomotion papers published in 2025-2026.
Community Size Proxies
ROS2 support is only as useful as the community that maintains it. Rough proxies:
- Largest research community: Unitree (G1+H1 combined ~10,000 units shipped, 50+ research labs)
- Most academic publications using the platform: PAL TALOS (15+ years, hundreds of papers), Unitree H1 (rapidly catching up)
- Most active GitHub orgs: unitreerobotics/unitree_ros2, pollen-robotics/reachy2-sdk, pal-robotics/talos_robot
- Fastest-growing Discord/forum activity: Booster Robotics (RoboCup-driven), Reachy (Hugging Face crossover)
- Largest closed-source community: Tesla β but it's not a robotics community, it's a fleet-AI userbase.
If your team's risk tolerance is low, picking a robot with thousands of deployed units and an active GitHub organisation removes whole categories of integration risk.
Setup Time Realities
A few sanity checks on "ROS2 ready" claims, based on community reports:
- PAL TALOS, Pollen Reachy 2: apt install + URDF, working sim in < 30 minutes.
- Unitree G1/H1: 1-3 hours including SDK build, dependencies, MJCF setup.
- Booster T1: 2-4 hours, mostly because Chinese-first docs need translation review.
- Atlas Electric, Digit: Onboarding is multi-day, NDA-gated, partner-program only.
- Tesla, Figure, 1X NEO: Not applicable β no public middleware integration path.
Use Case β Robot Mapping
For practical decision-making:
- Research lab needing maximum flexibility: Pollen Reachy 2 or PAL TALOS.
- Warehouse pick-and-place at scale: Agility Digit (best support) or Unitree H1/H2 (cheaper, requires more integration work).
- University teaching lab on a budget: Unitree G1 or Booster T1.
- Closed enterprise app with vendor-controlled AI: Tesla Optimus (if you can get one), Figure 02/03 (if you have BMW-scale procurement), 1X NEO.
- Simulation-only RL research: Unitree H1 in Isaac Lab. Free, ubiquitous, well-modeled.
Frequently asked questions
Does Tesla Optimus support ROS2?
No. Tesla's stack is fully proprietary FSD-derived silicon with no public SDK, no ROS or ROS 2 bridge, no Gym-style benchmark environment, and no third-party integration path. The only way to extend Optimus behaviour is to be Tesla.
Which humanoid robot has the best ROS2 support in 2026?
PAL TALOS is the gold standard β full ROS 2 Humble and Jazzy, official Gazebo simulation, MoveIt 2 configs, Nav2, and years of community validation. Pollen Reachy 2 is the open-source leader with full Hugging Face backing. Unitree G1 and H1 have the largest community after TALOS.
Can I run MoveIt 2 on Figure 02?
No. Figure 02 runs Figure's proprietary Helix vision-language-action system on dual NVIDIA RTX GPU modules. There is OpenAI API integration for higher-level prompts but no robotics middleware, no URDF, and no public SDK. The same applies to Figure 03.
What should I check before assuming a robot 'supports ROS2'?
Four things: URDF or MJCF availability (without a model, simulation is guesswork), MoveIt 2 config (essential for manipulation), NVIDIA Isaac Sim support (the industry standard for RL training), and community size (more users means more bug reports and faster fixes). 'ROS 2 compatible' from a vendor often means only the bridge, not the full stack.