- - For the Gasket Assembly, task, we present a learned diffusion policy and three procedural algorithms. - The procedural algorithms consist of gasket/channel detection, template matching and then actuation. The Gasket/Channel Detection box shows - gasket segmentation (above) and channel segmentation (below). The Template Matching box shows the three templates for the curved, straight and trapezoid - channels. The Straight/Curved Actuation box shows selection and actuation strategies for the straight and curved channels: (a) is Unidirectional insertion, - (b) is Binary search insertion, and (c) is Hybrid insertion. The numbers of the points on the channels represent the order and location where robot attempts - to place and press the gasket into. The arrows indicate the direction(s) of the slide(s). The half-points and quartile-points in binary insertion are labeled as - (2,3) and (4,6,7,5) respectively. In Hybrid insertion they’re labeled (4,5) and (6,8,9,5) respectively. For the trapezoid channel, we treat each segment of the - trapezoid as an instance of the straight channel. In the unidirectional approach (d) we process each segment in a counterclockwise manner, starting at the - blue segment. For hybrid and binary (e), we evaluate the blue segment, then the cyan segments, and finally the red segment. The learned policy proceeds - directly from the initial state to actuation (f). The Final State box shows the final assembled gasket. +
+ The Gasket/Channel Detection box shows gasket segmentation (above) and channel segmentation (below). + The Template Matching box shows the three templates for the curved, straight and trapezoid channel. + The Straight/Curved Actuation box shows selection and actuation strategies for the straight and curved + channels: (a) is Unidirectional insertion, (b) is Binary search insertion, and (c) is Binary+ insertion. + The colors on the channels represent the locations the robot attempts to place and press + the gasket into while the numbers represent the order they are placed and + pressed. Endpoints are green, midpoints are pink, half-points are blue and + the quartile-points are cyan. The arrows indicate the direction(s) of the + slide(s). For the trapezoid channel, we treat each segment of the trapezoid + as an instance of the straight channel. In the unidirectional approach (d) + we process each segment in a counterclockwise manner, starting at the blue + segment. For hybrid and binary (e), we evaluate the blue segment, then the + cyan segments, and finally the red segment. The learned policy proceeds + directly from the initial state to actuation (f). The Final State box shows the + final assembled gasket.
Automating Deformable Gasket AssemblyAbstract
- In Gasket Assembly, a deformable gasket must be
- aligned and pressed into a narrow channel. This task is common
- for sealing surfaces in the manufacturing of automobiles,
- appliances, electronics, and other products. Gasket Assembly is
- a long-horizon, high-precision task and the gasket must align
- with the channel and be fully pressed in to achieve a secure
- fit. We present and compare 4 methods for Gasket Assembly:
- one policy from deep imitation learning and three procedural
- algorithms. We evaluate each using 3D printed channels with
- 100 physical trials. Results suggest that deep imitation learning
- can fail (lowest quartile of alignment and insertion performance)
- on a straight channel in 2 of 10 trials, whereas a hybrid
- procedural algorithm achieves highest quartile performance in
- all 10 trials. The procedural algorithm also performs reliably
- on a curved channel but poorly on a closed trapezoidal channel.
+ In Gasket Assembly, a deformable gasket must be aligned
+ and pressed into a narrow channel. This task is
+ common for sealing surfaces in the manufacturing of automobiles,
+ appliances, electronics, and other products. Gasket
+ Assembly is a long-horizon, high-precision task and the gasket
+ must align with the channel and be fully pressed in to
+ achieve a secure fit. To compare approaches, we present 4
+ methods for Gasket Assembly:
+ one policy from deep imitation learning and three procedural algorithms.
+ We evaluate these methods with 100 physical trials. Results suggest that the
+ Binary+ algorithm succeeds in 10/10 on the straight channel
+ whereas the learned policy based on 250 human teleoperated
+ demonstrations succeeds in 8/10 trials and is significantly
+ slower.
@@ -280,10 +284,10 @@ Physical Trials & Results
10 for each procedural algorithm on each channel: straight, curved and trapezoid.
![]()
- 
-
- In Gasket Assembly, a deformable gasket must be - aligned and pressed into a narrow channel. This task is common - for sealing surfaces in the manufacturing of automobiles, - appliances, electronics, and other products. Gasket Assembly is - a long-horizon, high-precision task and the gasket must align - with the channel and be fully pressed in to achieve a secure - fit. We present and compare 4 methods for Gasket Assembly: - one policy from deep imitation learning and three procedural - algorithms. We evaluate each using 3D printed channels with - 100 physical trials. Results suggest that deep imitation learning - can fail (lowest quartile of alignment and insertion performance) - on a straight channel in 2 of 10 trials, whereas a hybrid - procedural algorithm achieves highest quartile performance in - all 10 trials. The procedural algorithm also performs reliably - on a curved channel but poorly on a closed trapezoidal channel. + In Gasket Assembly, a deformable gasket must be aligned + and pressed into a narrow channel. This task is + common for sealing surfaces in the manufacturing of automobiles, + appliances, electronics, and other products. Gasket + Assembly is a long-horizon, high-precision task and the gasket + must align with the channel and be fully pressed in to + achieve a secure fit. To compare approaches, we present 4 + methods for Gasket Assembly: + one policy from deep imitation learning and three procedural algorithms. + We evaluate these methods with 100 physical trials. Results suggest that the + Binary+ algorithm succeeds in 10/10 on the straight channel + whereas the learned policy based on 250 human teleoperated + demonstrations succeeds in 8/10 trials and is significantly + slower.