Sharpa brings D01, W02, and AE01 to IROS for tactile manipulation

What three products did Sharpa unveil?

Sharpa’s 28 September blog post, datelined Pittsburgh, introduces D01 as its first integrated tactile-sensing dexterous manipulation robot, W02 as a next-generation fully tactile compact hand following W01, and AE01 as a high-fidelity haptic exoskeleton data glove. The company frames the trio as covering whole-body task execution, fine contact, and demonstration capture.

A matching PR Newswire release the same day repeats the product names and IROS booth 514 at the David L. Lawrence Convention Center through 30 September. For another IROS humanoid thread already on AiLookout, see RoboParty’s RP1 open-source plans.

The company positions D01 as the whole-body tactile platform, W02 as the forearm-friendly successor to W01’s hand, and AE01 as the human-side capture/teleop glove that closes the data loop. That stack story is why the IROS materials read as a product family launch rather than a single arm announcement. Booth 514 at the David L. Lawrence Convention Center ran through 30 September 2026 EDT per the same posts.

Which D01 and W02 numbers are vendor-stated?

On D01, Sharpa’s materials cite an approximately 1:1 arm payload-to-weight ratio, maximum end-effector speed above 10.5 m/s, 1,000 Hz communication frequency, 0.2 mm repeatability, body electronic skin with 100 Hz tactile sampling, 0.1–20 N force range, and 0.2 N force resolution. The product page we opened also lists NVIDIA Jetson Thor as the main processor and ~40 kg weight without chassis. Those lines are Sharpa’s catalog claims.

On W02, the company says 21 active degrees of freedom, under 750 g mass, about 30% smaller than W01, fingertip Dynamic Tactile Array sensing from 5 mN to 30 N at 1 mm spatial resolution, palm/finger e-skin at 0.1–20 N, zero grasping gap for thin objects, and IP54 protection (IP67 with an official glove). AE01 is described with 22 encoders, 256 levels of fingertip haptic feedback, calibration-free use, and wired/wireless latency figures on the product page.

Sharpa’s D01 page also lists a dual-arm span with hands of 1478 mm, height range roughly 1109–1624 mm, 66 total degrees of freedom with hands, and Ethernet/Wi-Fi 6 interfaces. W02’s sheet adds 188 × 87.5 × 40 mm dimensions and a 3 Hz operating speed claim. AE01’s sheet lists size options S–XL, >3 h battery life, and wireless range up to 30 m. Keep these as catalog rows until a third party publishes a protocolled pick-and-place or teleop study.

Sharpa’s W02 narrative emphasizes fitting human-scale workspaces and reducing forearm load versus W01, with fingertip Dynamic Tactile Array sensing claimed from 5 mN to 30 N at 1 mm spatial resolution and a “zero grasping gap” claim for thin objects such as chopsticks. IP54 is the base environmental rating, with IP67 described when using an official glove. AE01’s haptic story is continuous fingertip vibration in 256 levels plus encoder-rich hand pose for egocentric demos—useful for teleop datasets, not proof of autonomous policy quality.

What should buyers separate from the booth claims?

IROS unveil posts mix architecture story and marketing demonstration. Speed, repeatability, and restaurant-style continuous-task anecdotes that appear in secondary write-ups need task definitions, lighting, and failure rates before they inform a purchase. Ask for the difference between teleoperated demos and autonomous policies, and for data rights when AE01 captures operator motion.

World-model and contact-rich learning papers on AiLookout—such as DreamTrue—are research baselines, not evaluations of Sharpa’s stack. Keep them in separate columns.

Ask for the evaluation protocol behind any restaurant or continuous-task anecdote: teleop versus autonomous, object set, hours of operation, and intervention rate. Tactile skin sampling rates and force ranges are catalog fields; they do not tell you whether the perception stack filters oil, gloves, or reflective packaging in your facility. For AE01, clarify who owns demonstration trajectories and whether operator biometrics are retained.

How does this relate to open humanoid platforms?

Sharpa’s posts describe proprietary products, not an open-source release schedule. That differs from RoboParty’s RP1 messaging about progressive open-sourcing. Teams comparing platforms should track license, simulators, and hand APIs explicitly rather than assuming IROS proximity implies the same openness.

Carry-forward note for this cycle: RoboParty’s RP1 open-source roadmap was already covered earlier today; AgentGarten’s renderer drop is likewise on the site. Sharpa is the additional IROS hardware stack that had not yet been written up. We still have not seen a new October code drop for RP1 beyond what that earlier article recorded.

If you are building a research stack, confirm simulator support, ROS/middleware bindings, and whether tactile arrays expose calibrated force images or only proprietary binaries. IROS proximity to open humanoid announcements does not transfer licenses. Keep Sharpa’s proprietary D01/W02/AE01 materials in a separate column from MIT/Apache robot code drops covered elsewhere on AiLookout.

What did we not test?

We did not attend IROS, handle D01/W02/AE01, measure tactile noise, or verify Jetson Thor integration. All performance figures are from Sharpa’s blog, PR, and product pages as opened on 10 October 2026.