We improved the platform and built a new version, BALLU2, which has a simpler and more robust design. The onboard computer newly adopted allows BALLU2 to be controlled by an algorithm. HC is the main contributor that designed the new iteration of BALLU, created the data-driven controller, conducted experiments, and wrote the manuscript. MA participated in devising new types of controllers for BALLU and writing the manuscript. DN and HN contributed to manufacturing and maintaining BALLU. DH is the inventor and the PI for BALLU, guided the project as the advisor, and contributed to the review of the paper. Funding Huang, C. et al. Miniaturized swimming soft robot with complex movement actuated and controlled by remote light signals. Scientific reports 5, 17414, https://doi.org/10.1038/srep17414 (2015). Hu FJ, Wu G (2020) Distributed error correction of EKF algorithm in multi-sensor fusion localization model. IEEE Access 8:93211–93218. https://doi.org/10.1109/ACCESS.2020.2995170 Le, V. H. et al. Shape memory alloy–based biopsy device for active locomotive intestinal capsule endoscope. Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine 229, 255–263, https://doi.org/10.1177/0954411915576946 (2015).

Schreuders, E. H., Grobbee, E. J., Spaander, M. C. W. & Kuipers, E. J. Advances in fecal tests for colorectal cancer screening. Current treatment options in gastroenterology 14, 152–162, https://doi.org/10.1007/s11938-016-0076-0 (2016). Time will tell if all companies that sell products built with artificial intelligence are scrupulous on this point. But in the meantime, it is worth noting that AIs demonstrating unexpected behaviours are by no means simply confined to research environments. They are already working their way into commercial products. Alazmani, A., Hood, A., Jayne, D., Neville, A. & Culmer, P. Quantitative assessment of colorectal morphology: Implications for robotic colonoscopy. Medical Engineering & Physics 38, 148–154, https://doi.org/10.1016/j.medengphy.2015.11.018 (2016).Madiouni R, Bouallègue S, Haggège J, Siarry P (2016) Robust RST control design based on multi-objective particle swarm optimization approach. Int J Control Autom Syst 14(6):1607–1617

All data for analysis were collected in simulation. Not only was nominal walking data from teleoperation collected, but walking sequences that include intentionally elongated and shortened double and single support phases as well as following recovery from such abnormal timings were also recorded.

ORIGINAL RESEARCH article

For verification of the proposed locomotion approach, straight walking of about 1.4m is tested with a desired walking speed of 0.18m/ s on the actual hardware. Because of the complexity and uncertainty of the system, rather than conducting formal nonlinear system analysis, BALLU’s walking performance is first qualitatively assessed by analyzing the pelvis’ trajectories and phase plot. Based on the careful analysis of the motions from experts’ teleoperation, a data-driven controller is suggested to handle the highly nonlinear dynamics. First, a set of teleoperation data in the simulation were collected in different scenarios. Next, to extract underlying state vectors of the complex dynamics, the correlations between the kinematic quantities were statistically evaluated; for each phase of walking, the quantities that have a high Spearman Correlation Score (SCC) were chosen as state variables. Then, their relationship was approximated using an artificial neural network, which is trained on the same data. Pourghodrat, A. et al. Disposable fluidic self-propelling robot for colonoscopy. Journal of Medical Devices 8, 030931, https://doi.org/10.1115/1.4027076 (2014).