cs.ROMar 18, 2026
SaveManufacturing Micro-Patterned Surfaces with Multi-Robot Systems
Organizations: Mechanical Engineering Department, Northwestern University
Abstract
Applying micro-patterns to surfaces has been shown to impart useful physical properties such as drag reduction and hydrophobicity. However, current manufacturing techniques cannot produce micro-patterned surfaces at scale due to high-cost machinery and inefficient coverage techniques such as raster-scanning. In this work, we use multiple robots, each equipped with a patterning tool, to manufacture these surfaces. To allow these robots to coordinate during the patterning task, we use the ergodic control algorithm, which specifies coverage objectives using distributions. We demonstrate that robots can divide complicated coverage objectives by communicating compressed representations of their trajectory history both in simulations and experimental trials. Further, we show that robot-produced patterning can lower the coefficient of friction of metallic surfaces. This work demonstrates that distributed multi-robot systems can coordinate to manufacture products that were previously unrealizable at scale.
Figures & tables
Fig. 1: Robot patterning. (a) Four robots patterning an image of a sculpture in the Balloon Dog series by Jeff Koons. From left to right: coverage objective, patterning by four robots without communication, patterning by four communicating robots. Communicating robots use the decentralized ergodic control algorithm, in which agents average their trajectory history to divide coverage. (b) From left to right: the club coverage objective, patterning by four non-communicating robots, patterning by robots sharing their trajectory history. (c) Robots patterning the club objective on acrylic.
Fig. 2: Robot design. The robot (a) uses an ESP32S3 microcontroller on a custom PCB. Angled wheels (b) minimize the footprint. The indentation tool moves from a low (c,e) to a high (d,f) position using a cam, which then releases and allows the tip (g) to impact the workpiece, forming a micro-dimple. (h) The robot moves according to the target density distribution to place these micro-dimples.
Fig. 3: Simulated results with and without communication. Four robots, each running the ergodic control algorithm, were simulated for 25 trials. No communication means that each robot is separately running ergodic control. Full communication means that agents average their trajectory history to divide the task, using the decentralized ergodic control algorithm. The black lines in (a) and (b) are the median values. (a) Agent trajectories are more different from each other when they communicate. (b) Performance is the average ergodic metric over the trial. Agents perform similarly across all cases, but perform worse on the more complicated balloon dog objective. (c) For each objective, the no communication case is on the left and the full communication case is on the right. Individual agent trajectories are different when agents communicate and the same when they do not communicate.
Fig. 4: Experimental results with and without communication. Note that these results include one experimental trial for each case. (a) Trajectory heterogeneity score for each objective. Communicating agents have a higher trajectory heterogeneity score than non-communicating agents. The heterogeneity scores were higher for the dog objective, which is more complicated. (b) The system’s ergodic metric over each trial. This shows a snapshot of the first 40 timesteps. The ergodic metric was minimized and stabilized to the shown values. Communication benefits the performance of the more complicated dog object, while it does not affect the performance of the club objective. (c) Images over time from the dog patterning experiment with 15 minutes total time. Elapsed time from left to right: 30 seconds, 2.5 minutes, 7.5 minutes, 15 minutes.
Fig. 5: Experimental trajectories for each robot. Individual robot trajectories for each case. Robots that communicate can divide the task, taking responsibility for a part of the coverage area. Robots that do not communicate have redundant coverage, each covering the whole area.
Fig. 6: Representative dimples on aluminum. A typical dimple after patterning (a) and a similar dimple (b) after being sanded to remove raised edges. Depthmap slices (c) show the excess material removed by sanding as well as the asymmetry due to robot motion during patterning.
Fig. 7: Friction testing setup. (a) Position history from robot and (b) resulting patterned workpiece with circular samples removed, (c) microscope images of samples taken from the high, medium, and low pattern density regions, (d) a sample bonded to its mount, (e) a reversed sample showing the leveling cone, and (f) the rotary tribometer setup showing the sample orientation, forces being measured, and input rotation.
| Duration [s] | Rotation Rates [rpm] |
|---|---|
| 30 | 400, 360, 320, 280, 240, 200 |
| 60 | 180, 160, 140, 120, 100, 90, 80, 70, 60, 50 |
| 90 | 45, 40, 35, 30, 25, 20 |
| 120 | 18, 16, 14, 12, 10, 8, 6, 4, 2 |
TABLE I: Rotary tribometer testing sequence.
Fig. 8: Rotary tribometer results. (a) Friction coefficient over test duration and (b) resulting Stribeck curves produced from the patterned workpiece. Low- and medium-density patterning reduces the onset of the mixed lubrication regime compared to a control, unpatterned surface. This result confirms the beneficial effect of robotic patterning for friction reduction.