Emerging data-driven approaches

Data-driven approaches have become increasingly relevant in robotic control as they have achieved unprecedented success in various applications, such as sliding contact control [1], agile and dynamic motion skills [2], and locomotion in the wild [3].

By leveraging large datasets of real-world robot motion or simulated data [1], data-driven approaches can learn complex control policies that effectively handle the nonlinear dynamics of the robot. Data-driven approaches can also be adaptable, enabling faster development and deployment of control policies for different tasks and environments. Therefore, data-driven approaches have the potential to improve the performance, robustness, and scalability of whole-body control for humanoid robots.

[1] N. Rudin, D. Hoeller, P. Reist, and M. Hutter, “Learning to walk in minutes using massively parallel deep reinforcement learning,” in Conference on Robot Learning. PMLR, 2022, pp. 91–100.

[2] J. Hwangbo, J. Lee, A. Dosovitskiy, D. Bellicoso, V. Tsounis, V. Koltun, and M. Hutter, “Learning agile and dynamic motor skills for legged robots,” Science Robotics, vol. 4, no. 26, p. eaau5872, 2019.

[3] T. Miki, J. Lee, J. Hwangbo, L. Wellhausen, V. Koltun, and M. Hutter, “Learning robust perceptive locomotion for quadrupedal robots in the wild,” Science Robotics, vol. 7, no. 62, p. eabk2822, 2022.