cs.ROOct 4, 2026
SaveTacGELLO: Tactile Contact Feedback for a 3D-Printed Teleoperation Leader
Organizations: University of Technology Sydney, NSW, Australia · Western Sydney University, NSW, Australia
Abstract
Passive leader arms transmit an operator's motion without returning remote contact or load cues. TacGELLO uses the existing servos of a 3D-printed leader for tactile contact feedback at the gripper trigger and current-based load cues at the arm joints. AnySkin contact detection activates a trigger position hold, while changes in follower joint current adjust leader-servo output limits to command resistance during lifting. Across six UR3 sessions with one operator, sent-command tracking error was 0.15--0.36 RMSE, with estimated trajectory lag of 82--98,ms. Tactile onset was logged in 60 of 62 segmented TacGELLO grasp intervals. Median commanded closure after tactile or width-based contact references was 0.7--2.1,mm with TacGELLO and 13.0--13.9,mm with the original interface.
Figures & tables
Figure 1: TacGELLO in use. The operator moves the 3D-printed leader and the UR3 follows it, here lifting a can with the tactile gripper. When the AnySkin fingertips detect contact, the leader’s trigger servo holds the trigger, so the operator feels the grasp.
Figure 2: TacGELLO hardware and example grasps. (a) The 3D-printed leader in the foreground and the UR3 follower behind it. (b) The UR3 fitted with the two-finger gripper. (c,d) The gripper holding a cylindrical container and a water bottle. Each fingertip carries an AnySkin tactile sensor. The photographs are cropped for presentation and illustrate the hardware only; they are not used to score experimental outcomes.
| System | Signal and presentation |
|---|---|
| GELLO [ 27 ] | passive remote leader |
| ALOHA [ 31 ] | passive remote leader |
| UMI [ 8 ] | handheld gripper; local mechanical interaction |
| Bi-ACT [ 7 ] | observer-based joint-force feedback |
| FACTR [ 19 ] | external joint torque to arm; gripper current to trigger |
| GELLO+F [ 24 ] | Franka external torque to current-controlled leader |
Table 1: Feedback source and presentation in selected interfaces. Local interaction in a handheld device differs from remote follower-force reflection.
Figure 3: Experimental system architecture. Black: command path. Blue: signals used for feedback. Feedback commands act through the leader’s own servos, by holding the trigger and adjusting joint servo output limits. The AnySkin USB streams are separate from the ESP32 motor-control link. No dedicated arm force/torque sensor or additional leader actuator is used.
| Object | Interface | Total | T | W | – |
|---|---|---|---|---|---|
| Chip | TacGELLO | 19 | 18 | 1 | 0 |
| Original | 23 | 0 | 21 | 2 | |
| Cup | TacGELLO | 20 | 19 | 1 | 0 |
| Original | 17 | 0 | 14 | 3 | |
| Bottle | TacGELLO | 23 | 23 | 0 | 0 |
| Original | 25 | 0 | 21 | 4 |
Table 2: Segmented intervals and available contact references. T: tactile onset; W: width fallback; –: no reference.
Figure 4: Sent joint commands and measured UR3 angles. (a) Three joints during the selected 20 s window with the most commanded travel. (b) Zoom on the fastest motion of the most active joint in that window. The displayed command has already undergone leader offset alignment and smoothing.
| Joint | Same-time | Lag | Aligned | |
| RMSE ( ∘ ) | (ms) | RMSE ( ∘ ) | ||
| Base | 0.18 | 84 | 0.014 | 3 |
| Shoulder | 0.19 | 98 | 0.017 | 3 |
| Elbow | 0.36 | 98 | 0.028 | 5 |
| Wrist 1 | 0.27 | 98 | 0.028 | 4 |
| Wrist 2 | 0.15 | 84 | 0.023 | 5 |
Table 3: Sent-command tracking over six sessions. Same-time RMSE pools samples; lag and aligned RMSE are medians over eligible joint-sessions ( ). Software rows use available session medians: six for arm stages and three for gripper stages.
Figure 5: Representative chip intervals selected near each interface’s median minimum commanded width, with pre-window commanded width at least 125 mm and segment duration under 6 s. Traces are aligned to 0.6 s before each segmented interval. The TacGELLO marker is logged tactile onset; the original-interface marker is its width-based contact reference. The AnySkin trace is the norm of baseline-relative logged field magnitudes and has no force calibration.
Figure 6: Per-interval command and encoder measurements for the original interface (grey) and TacGELLO (blue). Contact-dependent values use tactile onset where available and a width reference otherwise, including two TacGELLO fallbacks. Conditions also differ in command dynamics. Points include the large chip outlier.
| Object | Measure | Original | TacGELLO | |||
|---|---|---|---|---|---|---|
| Chip | Commanded closure (mm) | 13.0 [8.0, 24.3] | 21 | 0.7 [0.0, 1.5] | 19 | |
| Trigger rotation ( ∘ ) | 31.0 [28.7, 35.9] | 23 | 26.6 [25.8, 27.9] | 19 | ||
| Finger-servo closure ( ∘ ) | 27.2 [26.0, 31.7] | 23 | 23.4 [23.0, 24.2] | 19 | ||
| Cup | Commanded closure (mm) | 13.5 [6.7, 19.0] | 14 | 2.1 [1.4, 2.9] | 20 | |
| Trigger rotation ( ∘ ) | 20.0 [18.2, 23.0] | 17 | 16.9 [15.8, 18.3] | 20 | ||
| Finger-servo closure ( ∘ ) | 17.3 [15.6, 19.8] | 17 | 12.7 [11.7, 13.3] | 20 |
Table 4: Exploratory grasp outcomes: median [Q1, Q3]. Original: feedback off, rate-limited commands; TacGELLO: tactile hold, direct commands. Outcome-specific sample sizes are given in each row. is Cliff’s delta for TacGELLO minus original. Commanded closure uses different contact references across conditions.
| Cue | (A) | Peak limit | |
|---|---|---|---|
| 1 | 38 | 0.572 | 27 |
| 2 | 25 | 0.287 | 10 |
| 3 | 134 | 0.280 | 40 |
| 4 | 29 | 0.422 | 29 |
| 5 | 25 | 0.144 | 10 |
| 6 | 9 | 0.110 | 5 |
Table 5: Logged cue activations associated with the nominal 1200 mL bottle condition. counts valid active samples. is the median signed shoulder-current change relative to the empty-arm baseline. Peak limits are logged output-limit commands in raw servo units.