Project Description

Project PID2020-117509GB-I00 funded by MCIN/ AEI /10.13039/501100011033

Currently, robotics is experiencing a change in trend from specialized industrial robots, designed to perform repetitive operations on a routine basis, towards lighter and more versatile robots, increasingly integrated into our daily lives, sharing our familiar environments, and interacting with us. This change brings a new way of thinking about how robots should work. In an industrial setting, the tasks assigned to robots are perfectly defined and take place in a completely known and absolutely controlled environment. In this context, practically nothing is left to improvisation. In contrast, a robot operating in a human context lacks an exact model of the environment, which is only partially known and is subject to unexpected changes. Since the situation is unknown in advance, it is not possible to make a precise plan for the robots actions, so a margin of action must be left so that the robot can react appropriately to the current situation, behaving safely and efficiently.

Our departing hypothesis is that robots that operate in human environments must possess two specific qualities that we refer to as agility and gracefulness. By agility we understand the ability of the robot to rapidly change its course of action, which may involve changes in its speed, configuration, or mode of operation. Agility is crucial to respond in time to events that need a quick reaction. On the other hand, gracefulness is a desirable property for a robot that must interact with humans, since graceful behavior tends to avoid intense forces or sudden accelerations that could harm a human or the objects it manipulates. Gracefulness also tends to produce robotic behaviors that humans perceive as natural, thus increasing our confidence and ease of interaction with the robot.

In this project, we propose to formalize the concepts of agility and gracefulness in a quantitative way and to develop a trajectory optimizer capable of producing agile and graceful motions compatible with all the kinematic and dynamic constraints of the robot; that is to say, avoiding collisions and respecting joint bounds and limitations in the forces that the actuators can exert. Given an initial feasible trajectory, the optimizer has to improve it according to the selected cost function while still satisfying the aforementioned constraints. In particular, the proposed optimizer should be able to tackle tasks with (1) serial robots, (2) parallel robots and, in general, closed kinematic chains of any topology, and (3) fixed or mobile robots of any type manipulating a known load, all of them in environments with or without gravity.