humanoidrobots.systems

agibot/x1 · spec sheet

AgiBot Lingxi X1 specs

1.3 m open-source DIY humanoid whose hardware drawings, training code and inference stack AgiBot published on GitHub in October 2024.

Maker
Class
Humanoid
Status
Active
Availability
DIY build
Country
China, Shanghai
Level
L1 Compiled
01

Media

none on record
No licensed media on record for AgiBot Lingxi X1 yet. We only show our own renders from public URDF/MJCF, Wikimedia Commons files with their exact license, images from the robot’s own open repositories, and the maker’s official videos. No press images are hotlinked. Have an openly licensed image or an official video? Send it with the correction form.
02

Specifications

8/12 fields sourced
AgiBot Lingxi X1 specifications
Heightspecs.heightCm130cm
Massspecs.massKg33kg
Degrees of freedomspecs.dof31
Payloadspecs.payloadKg0.5kg
Top speedspecs.speedMps1m/s
Runtimespecs.runtimeMin120min
Battery capacityspecs.batteryWhunsourced
Pricespecs.priceUsdunsourced
Actuatorsspecs.actuatorsAgiBot PowerFlow joint modules (R86-2, R86-3, R52, L28)
Computespecs.computeunsourced
Sensorsspecs.sensorsunsourced
Year introducedyear2024
Licenselicense.hw / .sw / .docsSoftware: MulanPSL-2.0

Note · Speed of 3.6 km/h (~1.0 m/s) is from press. Hardware repo has dated releases through 2025-03-07. Repos have seen little activity since early/mid 2025 (train last pushed 2024-10), so treat X1 as a legacy open platform. AgiBot now publishes X2 URDFs instead (AgibotTech/agibot_x2_urdf). Licenses (GitHub license API, 2026-09-25): agibot_x1_infer is Mulan PSL v2; agibot_x1_hardware and agibot_x1_train have no license file, so the hardware files are all rights reserved and hw is left undeclared. DOF 31 = 29 PowerFlow joints (R86-2 x9, R86-3 x6, R52 x10, L28 x4) + 2 grippers, per the hardware BOM/SOP and agibot.com.cn; the earlier 34 had no upstream source. Walking policies ship as ONNX in agibot_x1_infer (rl_walk_leg.onnx, rl_walk_leg_shoulder.onnx). The training URDF/MJCF model the 12 leg joints only.

03

Openness

method
85 / 100Open hardware
ArtifactPublicWeightPoints
CADopenness.cadyes2525
Bill of materialsopenness.bomyes2020
Sim model (URDF/MJCF)openness.simyes2020
Firmwareopenness.firmwareno150
Softwareopenness.softwareyes1010
Trained policiesopenness.policiesyes1010
score = Σ weight × public85
04

Build package

v0.1.0 · full guide
BOM lines
50
Parts (qty)
1,363
Priced BOM
—
Fab parts
15
Print time
—
Material
—
Build steps
20
Est. assembly
—
Safety steps
8
Vendors
5

BOM by category · USD from priced lines

CategoryLinesQtyUSD
Power1012—
Cabling1016—
Structure1070—
Fasteners71,224—
Actuators531—
Electronics34—
Sensors23—
Other22—
Compute11—

Offers in CN from 5 vendors

Subsystems · assembly groups

GroupStepsMinSafety
Fabrication2——
Actuator pre-assembly2——
Arms2——
Legs2——
Waist1——
Electrical1—1
Torso2—2
Integration1—1
Bring-up3—3
Software3——
Shells1—1

Fabrication: CNC ×10 · SLA print ×4 · Sheet metal ×1

Firmware

Targets
Mini PC (x86, Ubuntu + PREEMPT_RT): agibot_x1_infer AimRT stack (rl_control_module, dcu_driver_module, joy_stick_module), Xyber-DCU v18 x2 over EtherCAT: xyber_controller SDK in agibot_x1_infer (src/module/dcu_driver_module/xyber_controller), PowerFlow R86-2/R86-3/R52/L28 actuators: ID and run mode are set over Type-C with the vendor host tool; actuator firmware is not published
Repo
github.com/AgibotTech/agibot_x1_infer
Calibrate
Mechanical zeroing with the toolkit jigs (waist zero blocks, foot zero blocks, 3x35 pins on leg joints, scribe lines on arm joints), then set each actuator's ID and run mode over Type-C in the order of the x1_id diagram, then remove all jigs. Joint-to-actuator bindings and PD gains are in src/install/linux/bin/cfg/x1_cfg.yaml (dcu_network, transmission, joint_conf).
Notes
Each DCU has an external IMU input and three CAN-FD buses (up to 8 actuators each) and runs at up to 1 kHz. The upper-body DCU drives each arm on its own CAN bus (5 rotary joints, 2 push rods, 1 gripper) plus the two parallel waist joints. The lower-body DCU drives the legs and the waist yaw. The DCU product link in the infer docs (agibotmall.com) no longer resolves.

Policies

  • walk, legs only (upper body held still), joystick velocity commands · ONNX Runtime in AimRT rl_control_module, 1 kHz control; trained with agibot_x1_train (Isaac Gym, humanoid-gym style PPO) · github.comLoaded by src/module/control_module/cfg/rl_x1.yaml (real) and rl_x1_sim.yaml (sim).
  • walk with shoulder pitch following the gait (walk_leg_arm mode) · ONNX Runtime in AimRT rl_control_module · github.comThe joystick doc says to use this mode on the real robot, because the models were updated iteratively.
  • x1_dh_stand training recipe (standing/walking, 12 leg actions) · Isaac Gym Preview 4 + rsl_rl-style PPO (legged_gym / humanoid-gym lineage) · github.comNo trained checkpoint in the repo; export_onnx_dh.py produces the ONNX that the infer stack loads.

Caveats · stated by the package

  • LICENSING: agibot_x1_hardware and agibot_x1_train have no LICENSE file, so all rights are reserved by default. You may read and link the drawings, BOMs, SOP, URDF/MJCF and training code, but building from, modifying or redistributing those files is not licensed. Ask AgiBot for permission before selling kits or republishing CAD. Only agibot_x1_infer is open-licensed (Mulan PSL v2, MulanPSL-2.0).
  • No upstream document has a price. The BOMs and toolkit list give quantities, materials and reference Taobao/Tmall/JD links only, so every line is unpriced. The AgiBot store (store.agibot.com.cn) is a JS app, and no X1 part listing or price could be verified there.
  • Actuator counts conflict. The 2025-03-07 total BOM lists R86-2 x6, but the SOP says 9 R86-2 kits are needed, the 2024-10-24 BOM lists R86-2 x9, and the total BOM itself has 9 four-point-contact bearings and 9 bearing-ring sets (one per R86-2 kit). This package uses 9. That gives 29 joints (R86-2 x9, R86-3 x6, R52 x10, L28 x4) plus 2 grippers = 31, matching the 31 DoF on agibot.com.cn; the registry's 34 DoF is not backed by any upstream source.
  • The per-joint actuator assignment (which PowerFlow model drives which leg and waist joint) is only shown in SOP photos and the x1_id diagram, not in text. Check it against the photos before ordering.
  • Material/process conflicts between BOM revisions. The 2024-10-24 BOM specifies painted 3D-printed resin shells, while the 2025 total BOM says ABS (SLA is assumed here). The shank shells (左侧小腿-1/2 and mirror) and the waist PCB bracket are 6061 anodized in 2024 but ABS in 2025. The wrist cross shaft is 40Cr in 2024 but 17-4PH in 2025. Several fastener and pin quantities differ between the 2025-01-08 and 2025-03-07 BOMs; the 2025-03-07 total BOM is used.
  • Custom parts without drawings: the ankle and waist linkage assemblies (6QUUU1-4), the rod ends (6FUU4K/6FUUBQ, PDF only) and the spherical bearings (6FUU4J) are AgiBot custom parts with no vendor. They are presumably in the structural pack.
  • Battery spec is inconsistent: the 2024 BOM asks for a 48 V 8 Ah pack, '12S preferred', but the toolkit charger is 13S 54.6 V. Confirm pack chemistry and cell count before charging. The BOM gives no BMS or fuse.
  • The router (GL.iNet MT2500A), the 5 V module, the 48-to-24 V wrist converters and the itemised cable list come only from the 2024-10-24 BOM. The 2025 total BOM drops them (the router bracket remains), so they may be obsolete.
  • The assembly SOP is a Chinese .docx inside a zip, with about 640 photos; left/right differences are only shown in photos. The assembly and desk-bring-up videos are only on Baidu Netdisk (links with extraction codes in the repo's .txt files), which is hard to reach outside China.
  • Actuator firmware and the Type-C host tool used for ID and run-mode setup are not published. The DCU product page linked in the infer docs (agibotmall.com) no longer resolves.
  • The training URDF/MJCF in agibot_x1_train models only the 12 leg joints; arms, waist and grippers are fixed. There is no whole-body sim model. The train repo was last updated on 2024-10-23, and the hardware has changed twice since.
  • The hardware repo has no README, and the large SolidWorks/STEP archives are split multi-part zips (.zip.001-.007, about 100 MB each). You need the whole set to extract them.

Package agibot/x1@0.1.0 · upstream agibot_x1_hardware@29e662154dec89e07626ab6d30e5de11baeae439 (2025-04-18; drops 20241024, 20250108, 20250307: total BOM, structural BOM, toolkit list, assembly SOP .docx, 2D PDF drawings, per-part STEP, SolidWorks 2022 sources); agibot_x1_infer@9e0b818804d644fb9c9663e932dd33b03b24dfa4 (2025-04-03); agibot_x1_train@e6651b9ab843fd1b1be70c087bfb7e8b28e44ccd (2024-10-23). · 18 package sources

05

Models, repositories, links

6 links
06

News and papers

0 stories · 0 papers

News mentioning AgiBot Lingxi X1

The news feed is not reachable right now; this page regenerates every 30 minutes.

Papers using AgiBot Lingxi X1

The papers feed is not reachable right now; this page regenerates every 30 minutes.
07

Sources

last checked 2026-09-25
  1. [1]agibot_x1_infer: ONNX walking policiesgithub.com/AgibotTech/agibot_x1_infer/tree/main/src/module/control_module/policychecked 2026-09-25
  2. [2]agibot_x1_train: x1.urdf and xyber_x1_flat.xmlgithub.com/AgibotTech/agibot_x1_train/tree/main/resources/robots/x1checked 2026-09-25
  3. [3]agibot_x1_hardware 2025-03-07 BOM and SOP (joint count)github.com/AgibotTech/agibot_x1_hardware/tree/main/%E6%99%BA%E5%85%83%E7%81%B5%E7%8A%80X1_20250307checked 2026-09-25
  4. [4]Wikipedia: AgiBoten.wikipedia.org/wiki/AgiBotchecked 2026-09-24
  5. [5]agibot_x1_hardware repogithub.com/AgibotTech/agibot_x1_hardwarechecked 2026-09-24
  6. [6]Humanoid.guide AgiBot X1humanoid.guide/product/agibot-x1/checked 2026-09-24
08

Suggest a correction

Corrections need a public source. They go to the Hyperspawn intake queue and are reviewed before the record changes; the page updates on the next data release.

now: 130
POST hyperspawn.io/api/v1/reports
09

Related

10

Cite this page

humanoidrobots.systems. “AgiBot Lingxi X1 specs.” Hyperspawn, data revision 2026-10-08. https://humanoidrobots.systems/agibot/x1

Machine-readable: /data.json (record agibot/x1).