<- Back to news

DA-GRD uses sparse touch to recover a grasp after the object moves

A*STAR, NTU and Tsinghua researchers stop tactile exploration when the remaining object poses support the same grasp. Real hardware reaches 71.7% lift success, but only 38.3% task-conditioned success.

Preprint · arXiv v1 · simulation and real-robot evaluation · no public code verifiedSource date: Read the primary source ↗
active tactile explorationgrasp recoverytouch-based localizationrobot manipulation
Diagram showing a stale visual grasp, sparse tactile probes and a recovered robot grasp in the DA-GRD pipeline.
Original RoboSkin.ai schematic of DA-GRD belief updating and grasp-aware stopping. It is explanatory artwork, not an experimental image.

Researchers from A*STAR's Institute of Advanced Intelligence and Computing, Nanyang Technological University and Tsinghua University released DA-GRD on September 24, 2026. Decision-Aware Grasp-Relevant Disambiguation addresses a specific failure: vision selects a grasp, the object moves before execution, and the robot must recover without another camera observation. The method probes with touch until the remaining object-pose hypotheses support a common executable grasp. Paper and version record.

Explore tactile datasets for the broader comparison.

Key takeaways

  • In MuJoCo, the principal DA-GRD variant reaches 84.7% physical lift success across ten objects, compared with 63.7% for a fixed 15-scan baseline using the same planar belief representation.
  • The mean of the ten per-object probe averages on lift-success episodes is 4.13, 72.5% below a fixed 15-probe budget. Initial-grasp successes count as zero additional probes; this is not an all-attempt time saving.
  • On a UR5, 60 trials across six objects produce 71.7% lift success but 38.3% task-conditioned success. The gap shows that lifting an object is easier than recovering the originally intended grasp.

What changed

DA-GRD does not try to reconstruct a complete object pose before acting. It starts from the remembered RGB-D point cloud and task-conditioned grasp produced by a visual-language and AnyGrasp pipeline. When that grasp becomes stale, contact or free-space observations update a weighted 256-particle belief over planar translation and yaw.

Candidate probes approach vertically or horizontally. The selector evaluates a counterfactual miss and surface novelty, balanced against motion and repeat costs; it does not calculate a full expectation over every possible hit and miss outcome. A grasp-aware stopping rule executes once enough of the belief mass supports a common grasp. This is different from fixed tactile scanning, which spends the same interaction budget whether or not the next touch changes the decision.

Results under the reported conditions

The simulation applies translations up to plus or minus 5 cm on each planar axis and yaw changes up to plus or minus 45 degrees to ten rigid objects. Each object has 100 shared perturbation episodes per method. A lift counts as physically successful when the robot raises the target by 5 cm. DA-GRD's main policy reaches 84.7%, versus 9.1% for the unrecovered stale grasp, 21.2% for the original XY-only scan baseline, and 63.7% after that baseline receives the same SE(2) belief. Full evaluation.

The stricter task-conditioned measure asks whether the final grasp is also consistent with the grasp originally selected for the task. On that measure, DA-GRD reaches 57.3%, while the adapted fixed-scan baseline reaches 49.2%. A pose-sensitive DA-GRD variant reaches a similar 57.6% but uses 7.70 probes on average, compared with 4.13 for grasp-aware stopping.

The physical setup uses a UR5, a Robotiq 2F-140 gripper, two finger-mounted force-sensitive resistors for lateral contact and a wrist FT300 sensor for vertical contact. Across ten trials on each of six objects, lift success ranges from 60% to 80%; task-conditioned success ranges from 30% to 50%. The overall averages are 71.7% and 38.3%, with a reported mean of 4.20 probes across the six per-object averages. Probe count alone does not include the duration of each contact, motion or reset.

What this means for robotics

RoboSkin analysis: the useful idea is that touch should resolve the action, not necessarily the entire scene. If multiple object poses all imply an acceptable grasp, further localization spends time and risks moving the object without improving the decision. That principle could reduce the interaction cost of tactile manipulation systems that operate after occlusion or scene change.

The real-world result also argues for reporting more than a binary lift score. A controller can pick an object up while missing the requested handle, orientation or functional region. Keeping physical and task-conditioned success separate makes contact-aware recovery easier to evaluate honestly.

Limitations and availability

DA-GRD is an arXiv v1 preprint, and RoboSkin.ai has not reproduced it. The belief only covers planar SE(2) motion. Complex non-planar shapes can produce different three-dimensional contacts that look identical after projection; partial remembered geometry can preserve the same ambiguity. The physical implementation also uses both fingertip contact switches and a wrist force-torque sensor, so “sparse touch” does not mean sensor-free deployment.

The November 4 review of the paper and arXiv record did not verify an official project repository, dataset, checkpoint or software license. The arXiv record provides the manuscript under its selected submission license, which does not by itself release implementation code.

A stopping threshold is not a calibrated success probability

DA-GRD first attempts the remembered grasp. If it succeeds, the episode needs no additional exploration, but it has still made physical contact. After failure, a geometric belief over planar poses guides probing and stopping. The reported 70% grasp-coverage threshold concerns posterior mass compatible with a candidate grasp; it does not mean that 70% of physical attempts will succeed. Sensor errors, incomplete geometry and motion caused by probing can break that interpretation. Decision rule and protocol.

Report separatelyAvoid substituting
Initial grasp, added probes and final grasp“Zero probes” for contact-free execution
Lift success and intended grasp consistencyAny pickup for task completion
Per-object successful-episode probe meanAn average over all attempted episodes or wall-clock cost
Geometric posterior coverageCalibrated confidence in retention, damage avoidance or task success

For a deployment comparison, keep failures, timeouts, probe duration, recovery motion and interventions in the same attempted-episode ledger. A faster successful subset can coexist with expensive failures. The paper’s fixed-scan and decision-aware methods also differ in selection, refinement and stopping, so the aggregate gap does not isolate stopping alone.

This suggested evaluation extends the manipulation decision map. Preserve planar assumptions and contact-sensor calibration when adapting the method; a lower uncertainty over projected poses does not recover unobserved height, tilt or contact normals.

*Source review updated November 4, 2026. Original publication dates are retained. Results belong to the cited authors; RoboSkin has not run these systems or downloaded their experimental payloads.*

Continue the topic

Adaptive robot graspingContinual 6-DoF grasping learns from outcomes without weight updatesCross-embodiment graspingFunCo-Grasp transfers grasp structure across different robot handsProprioceptive dexterityA blind grasp reflex lets a dexterous hand feel through proprioception