Project Description

Project PID2024-161931NB-I00 funded by MCIN/ AEI /10.13039/501100011033 and by ERDF, UE

This project focuses on understanding, generating, and executing agile robotic movements. Agility can be considered as the ability of a creature to move quickly and easily, that is, with elegant dynamic motion and resource optimization. Recently, a small number of groups worldwide have started to showcase robots with agile movements: Boston Dynamics Atlas robot, autonomous racing drones, or Deep Robotics quadrupeds with impressive abilities. Unlike these examples, at ALPHA, we are particularly interested in hybrid scenarios. These hybrid scenarios can be understood in different ways: hybrid locomotion consisting of flight and contact (jumping and flying), hybrid locomotion with manipulation (bouncing or intercepting objects of similar mass in flight), or hybrid movement modes consisting of combining two movements of different natures, such as take-off (running, then flying) or landing (flying, then running, then stopping). In biology, these hybrid maneuvers occur very quickly and are impressively efficient at precisely controlling motion, often in critical situations. Generally, they give the creatures that master them a key evolutionary advantage. In robotics, such efforts are very
challenging and require a holistic approach that encompasses state estimation, control, skill learning, and advanced robot design.

To conduct this research, we propose starting with a hybrid robot prototype that we designed and built during our latest EBCON project. This prototype consists of a flying platform equipped with one or more articulated limbs. These limbs can be used to exert forces against the environment or other objects, playing the role of a leg or arm. They can also assist in movement while flying, taking on the role of a tail. In general, this robot constitutes a highly capable tool for exploring hybrid agile movement, which we will develop along four main axes.

The first axis is the robot itself, which must be lightweight, powerful, compliant, and driven by force and torque. The second axis is perception and state estimation, where the challenges stem from the need to achieve very fast and accurate estimates of a rich set of variables, ranging from robot dynamics to environmental features or even the dynamics of other objects with which the robot might interact. The third axis is control, which must be predictive, optimal, and account for the dynamics of its multijointed body. The fourth and final axis is the need to train the robot for complex movement skills, requiring machine learning tools and a subsequent skills database that can be consulted as needed according to the movement plan. Overall, the ALPHA project will contribute to the understanding and generation of very complex movements in hybrid locomotion systems. We believe that mastering this type of movement will be key for modern robotics in the near future. To this end, this project explores the most demanding scenarios. The knowledge and insights gained from this research will provide enormous value to the robotics community.