OpenArm

Enactic's open-source 7-DoF humanoid arm and officially supported LeRobot bimanual manipulation platform

OpenArm

One-line judgment

OpenArm is the strongest first-tier main experiment platform candidate for bouquet-making VLA work. Official LeRobot support, bimanual configuration, 7-DoF reach, and a low BOM line up well, but final performance depends on the end-effector and sensing stack.

Overview

OpenArm is an open-source 7-DoF humanoid robot arm led by Enactic in Japan. It is listed in the official Hugging Face LeRobot robot documentation and exposes single-arm and bimanual leader-follower workflows through openarm_follower, openarm_leader, bi_openarm_follower, and bi_openarm_leader.

ItemDetails
LeadEnactic
Type7-DoF humanoid robot arm
LeRobot supportOfficial docs for single-arm/bimanual teleoperation and dataset recording
ActuationDamiao QDD motors, CAN/CAN-FD
Payload4.1kg nominal, 6.0kg peak per arm before end-effector weight
PriceAbout $6,500 BOM for a complete bimanual system
OpennessCAD, URDF/xacro, CAN control, ROS 2, teleoperation repos
Main useContact-rich manipulation, bimanual data collection, imitation learning, VLA policy experiments

Fit for Flower Handling and Bouquet Making

OpenArm is one of the most interesting purchase candidates among officially supported LeRobot platforms. It has much more reach and payload than SO-101, is much cheaper than Reachy 2 or industrial dual-arm cells, and the bimanual setup is exposed directly through LeRobot commands.

CriterionAssessment
Bimanual workStrong fit. bi_openarm_follower is documented for bimanual leader-follower control.
Reach for floral tasksStrong fit. Human-scale arm proportions make tabletop stem alignment, wrapping, and handoff tasks practical.
DexterityThe arm is capable, but delicate petals/stems will depend on the end-effector. Consider soft pads, LEAP/Allegro, or Robotiq-style grippers.
DurabilityThe project describes aluminum and stainless steel structural parts for repeated data collection.
PriceAround $6.5k BOM for bimanual hardware is very aggressive for research-grade dual-arm manipulation.
CustomizationVery high. Hardware files, simulation, ROS 2, and CAN control are open.
RiskNewer ecosystem than ALOHA or Franka. Confirm assembled-unit availability and lead time before purchase.

Recommended role: first-tier experimental platform for bouquet-making VLA data collection.
Main caveat: delicate grasping quality will depend heavily on gripper choice and tactile/vision setup.


LeRobot Integration

LeRobot documents both single-arm and bimanual leader-follower setups for OpenArm. The important part is the workflow, not copying long command blocks into the hardware note.

FlowMeaning
openarm_follower / openarm_leaderValidate the camera-action data loop with one arm
bi_openarm_follower / bi_openarm_leaderCollect bimanual demonstrations for bouquet work
lerobot-recordSave datasets in the Hugging Face Hub flow for ACT, SmolVLA, pi0-style, or custom VLA experiments

Use the LeRobot OpenArm docs for exact commands, since command-line flags can change across versions.


Hardware and Software Stack

ComponentDetails
DoF7-DoF per arm
CommunicationCAN/CAN-FD
ControlPosition control, gravity compensation, teleoperation, force-feedback-oriented design
SimulationMuJoCo and Isaac Sim/Isaac Lab repositories
ROSROS 2 packages available
Hardware licenseOpenArm hardware repo uses CERN-OHL-S-2.0
Software licenseMain software repos use Apache 2.0

OpenArm is not just a low-cost arm. Its value is that it is an open hardware stack shaped around VLA data collection. With a bimanual leader setup, human demonstrations for flower trimming and bouquet assembly can be collected in the same LeRobot pipeline used for ACT, SmolVLA, pi0-style policies, and future custom VLA inference.


Purchase Checklist

  1. Confirm assembled or DIY kit supplier and lead time.
  2. Prepare CAN-FD adapters, power, emergency stop, and table mounting.
  3. Keep any cutting tool tests inside a physical safety cell.
  4. Start with the default gripper, then switch to soft pads, compliant grippers, or dexterous hands if stem damage is high.
  5. Validate the single-arm openarm_follower pipeline before expanding to bi_openarm_follower.

References


See Also