Kinodynamic planning of efficient and agile robot motions
Total activity: 2
Type of activity
Ministerio de Ciencia e Innovación
Funding entity code
Robotics is flourishing. Innovative robot mechanisms constantly see the light of day, and their use may increase dramatically in the near future. Whether on Earth or in Space, from research labs, to medicine, or the industry, we see parallel and walking robots, flying manipulators, anthropomorphic hands and arms, humanoids, and other sophisticated machines in action. The capacity to autonomously plan and perform complex motions is key on such devices, and robotics has provided many solutions to this end. Despite the impressive advances, however, roboticists are beginning to recognize that robots still move far too conservatively (R. Tedrake. "Underactuated robotics." MIT Open Course Ware. http://underactuated. csail.mit.edu), and accomplish only a fraction of the tasks and achieve only a fraction of the performance that they are mechanically capable of. This must be attributed to the fact that many robots are fundamentally limited by control technology that matured on rigid robotic arms in factory environments. Such robots use high-gain control loops, and therefore considerable joint torque, to cancel out their natural dynamics to strictly follow a desired trajectory. This approach to robot motion makes the problem tractable, but comes at a high price: a robot consumes much more energy than a human does to perform the same task, and it requires an oversized structure to support excessively large motors and resist their reactions. The result is a machine that is much less efficient and agile when compared to what a human, or an animal, would be in accomplishing a similar task.
The objective of this project is to investigate how energy-efficient and agile robot motions can be planned and executed in an efficient and reliable way. While robot movements are usually rigid and stereotyped, our aim is to make them more graceful. This does not mean to avoid jagged movements by simply smoothing the trajectory, but to adapt each movement to the natural frequency of the robot parts and manipulated objects, taking advantage of gravity, inertia, and centripetal forces, and thus reducing the internal forces and global effort of the robot.
A departing hypothesis is the realisation that such motions can only be generated by (1) taking the full robot dynamics into account, and (2) making an optimal use of the limited power, energy, and strength capacities of the robot equipment. To a large extent, this calls for offloading lower-level control loops in their task to achieve feasible, conservative motions, transferring part of their duty to higher-level motion planners that, by considering the full robot dynamics, are able to achieve graceful natural motions compliant with motor torque, energy storage, or material resistance limitations. A second hypothesis is the observation that there are new computational tools from motion planning, numerical continuation, differential geometry, multibody dynamics, and robot singularity theory, that can be employed to devise a high-level motion planner taking all such limitations into account.