ARTS: An Open-Source, Reproducible Three-Arm Platform for Autonomy Research in Minimally Invasive Surgery
This preprint presents ARTS, an open-source surgical robotics research platform with three arms and a USD 137,320 bill of materials. Real-hardware tests measured ISO 9283 pose repeatability of 0.14 to 0.25 mm, while three expert operators completed a laparoscopic peg-transfer task. Purchasable components and synchronized recordings could help laboratories collect comparable data for surgical robot learning.
Can purchasable components support a reproducible surgical research platform that combines wristed instruments, energy delivery, robotic stereo vision and synchronized data collection with measured control performance?
Surgical robot learning lacks sufficiently shared, comparable data. Laboratories often use clinical systems with different interfaces and restricted access, making it difficult to reproduce hardware, access raw endoscopic images and align recordings with robot actions.
The da Vinci Research Kit (dVRK) depends on donated, retired clinical hardware. MOPS provides wristed instruments on industrial arms, the Rodriguez et al. platform uses non-wristed instruments, and DASH adapts da Vinci instruments to other robots. The authors identify a missing combination: robotic stereo endoscopy, wristed instruments with energy delivery and hardware-synchronized stereo recording in one purchasable system.
Two UR5e arms drive da Vinci X/Xi instruments, while a UR5 carries a da Vinci Si stereo endoscope coupled to two FLIR cameras. ROS 2 Jazzy on Ubuntu 24.04 uses pinned dependencies and shared installation scripts for bare metal and containers. Teleoperation, scripted motion and learned policies share a tool-tip target interface; quadratic programming enforces remote-center-of-motion (RCM) and joint constraints. Instrument arms run at 500 Hz and the endoscope arm at 125 Hz. A synchronization board measures camera exposure timing, and energy activations enter the recordings. A software-independent hardware stop holds all three arms. Hardware designs, firmware, software and the BOM are released under Apache 2.0.
All experiments used real arms. ISO 9283 testing, five poses visited 30 times each, gave ARTS repeatability of 0.14 to 0.25 mm versus 0.09 to 1.24 mm for dVRK Classic. Three 5 min trials measured physical RCM mean errors of 0.34 to 0.37 mm, maximum 1.09 mm. Over 30.5 min, each camera recorded all 54827 trigger pulses; stereo skew was 2.7 ± 3.2 µs, maximum 11 µs. Hand-to-arm latency was 58 ms and 54 ms; switching from servoj to speedj reduced the compared latency from 84 ms to 58 ms. Camera-to-monitor latency was 68 ms. A 10 min control run reached 499.99 Hz with 3.8% jitter and no missed cycles after startup. After 5 min of practice, three experts completed FLS peg transfer in 93.2 to 144.8 s, mean 121.5 s, without drops. Two of three novices finished in 162.1 s and 219.1 s; the third transferred five of six triangles by 300 s and dropped one. During peg transfer, median kinematic RCM deviation was 0.022 to 0.034 mm per arm and trial. One expert demonstrated suturing qualitatively.
No autonomous task performance or clinical benefit was evaluated. Physical RCM tests used free air without a cannula. ARTS repeatability testing had RCM disabled, while dVRK retained its cannula. The 10 min timing run used an earlier controller without RCM. The operator study was small; one trial interrupted by a haptic device failure was excluded and replaced. Instrument zeros differ by a few degrees across instruments. Jaw input updates about every 90 ms in 4.7° steps. HMD video latency and the head-presence sensor's departure delay were not measured. No independent group has rebuilt the system. The BOM uses used clinical equipment and prices the discontinued UR5 CB3 as a UR5e.
Surgical robotics developers and teams building demonstration datasets could use the released platform now for research into teleoperation and learning. Broader adoption depends on an independent rebuild, more consistent instrument calibration and testing under tissue and trocar loads. The USD 137,320 BOM provides a concrete equipment budget, subject to its used-equipment assumptions. A timetable for clinical deployment is not reported in the paper.
The full text is not republished here because the paper's license does not allow it. Read the original on arXiv.