Rapid co-design of Buoyancy-assisted robots for Challenging Locomotion using Gaussian Evolutionary Specialists
Rapid co-design of Buoyancy-assisted robots for Challenging Locomotion using Gaussian Evolutionary Specialists
This paper introduces Gaussian Evolutionary Specialists (GES), a reinforcement learning-based framework for co-designing legged robots. It addresses behavioral diversity collapse by maintaining specialist policies that adapt to different morphologies. Evaluated on the Buoyancy-Assisted Light Legged Unit (BALLU) robot, GES achieves superior performance in obstacle traversal and ramp walking compared to baseline methods, reducing co-design time by 67.4% for the same number of evaluations. Hardware experiments confirm that optimized designs can traverse obstacles up to 24 cm high, significantly improving upon the original design's capability.
The study presents a reinforcement learning framework called Gaussian Evolutionary Specialists (GES) that optimizes both morphology and control for legged robots, demonstrating improved obstacle traversal and ramp walking performance on the Buoyancy-Assisted Light Legged Unit (BALLU) robot.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Control
- TeleoperatedOpen source S5
Reported numbers
obstacle height gain
5–9
View original evidence
Light Legged Unit (BALLU) robot for obstacle traversal and ramp walking over design spaces of up to four dimensions. GES achieves higher best design performance than Monolithic and Mixture-of-Experts baselines, with gains of 5.0-9.0 cm in obstacle height and 10.0-12.7 points in ramp grade. For the same 200 design-optimization evaluations, GES reduces wall-clock co-design time by 67.4% compared with retraining a poli
Open source S4ramp grade gain
10–12.7
View original evidence
Light Legged Unit (BALLU) robot for obstacle traversal and ramp walking over design spaces of up to four dimensions. GES achieves higher best design performance than Monolithic and Mixture-of-Experts baselines, with gains of 5.0-9.0 cm in obstacle height and 10.0-12.7 points in ramp grade. For the same 200 design-optimization evaluations, GES reduces wall-clock co-design time by 67.4% compared with retraining a poli
Open source S4obstacle traversal height
≤24
View original evidence
cy for every candidate. Finally, teleoperated hardware experiments evaluate the optimized BALLU morphology, which traverses obstacles of up to 24 cm, compared with 8 cm on the original design.
Open source S5original design obstacle height
8
View original evidence
cy for every candidate. Finally, teleoperated hardware experiments evaluate the optimized BALLU morphology, which traverses obstacles of up to 24 cm, compared with 8 cm on the original design.
Open source S5design-optimization evaluations
200
View original evidence
Light Legged Unit (BALLU) robot for obstacle traversal and ramp walking over design spaces of up to four dimensions. GES achieves higher best design performance than Monolithic and Mixture-of-Experts baselines, with gains of 5.0-9.0 cm in obstacle height and 10.0-12.7 points in ramp grade. For the same 200 design-optimization evaluations, GES reduces wall-clock co-design time by 67.4% compared with retraining a poli
Open source S4ramp grade gain upper bound
10–12.7
View original evidence
Light Legged Unit (BALLU) robot for obstacle traversal and ramp walking over design spaces of up to four dimensions. GES achieves higher best design performance than Monolithic and Mixture-of-Experts baselines, with gains of 5.0-9.0 cm in obstacle height and 10.0-12.7 points in ramp grade. For the same 200 design-optimization evaluations, GES reduces wall-clock co-design time by 67.4% compared with retraining a poli
Open source S4obstacle height gain upper bound
5–9
View original evidence
Light Legged Unit (BALLU) robot for obstacle traversal and ramp walking over design spaces of up to four dimensions. GES achieves higher best design performance than Monolithic and Mixture-of-Experts baselines, with gains of 5.0-9.0 cm in obstacle height and 10.0-12.7 points in ramp grade. For the same 200 design-optimization evaluations, GES reduces wall-clock co-design time by 67.4% compared with retraining a poli
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
cy for every candidate. Finally, teleoperated hardware experiments evaluate the optimized BALLU morphology, which traverses obstacles of up to 24 cm, compared with 8 cm on the original design.
Open source S5
Light Legged Unit (BALLU) robot for obstacle traversal and ramp walking over design spaces of up to four dimensions. GES achieves higher best design performance than Monolithic and Mixture-of-Experts baselines, with gains of 5.0-9.0 cm in obstacle height and 10.0-12.7 points in ramp grade. For the same 200 design-optimization evaluations, GES reduces wall-clock co-design time by 67.4% compared with retraining a poli
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org