Periodic Report Summary - FLEXIBLEROBOTBEHAV (Flexible behaviours for humanoid robots and digital humans)
Recent economic studies from the United Nations Economic Commission for Europe or from the Japan Robot Association indicate that robotics is expected to become a major industry in Europe and in the rest of the world in the coming decades. Obviously, most of this growth is still expected to take place in classical long-standing applications of robots such as manipulator robots dedicated to repetitive well-defined tasks. But a second class of applications is expected to emerge in the forthcoming years, with more autonomous robots undertaking more complex interactions with less controlled environments, service robotics. Of course, before such applications have any serious chance to materialise, many complex scientific and technical problems still need to be solved in actuation and sensor technology, in vision, mechanical design, control theory and computer science.
This is where research on humanoid robots becomes interesting. It is not clear today whether a humanoid shape might be of interest when designing a robot, and for which applications. But a clear interest of humanoid robots today is that they concentrate in one single research platform most of the scientific and technological challenges that emerge when considering 'complex interactions with uncontrolled environments'. And while appearing first as bare integrations of technological and scientific achievements developed formerly and independently, these platforms reach now a level of maturity that turns them into new driving forces for robotics research, bringing up whole new problems and challenges of unprecedented complexity and technical difficulty.
The main objective of this research and training project is to enhance the algorithms and control laws of existing humanoid robots in order to obtain a walking behaviour versatile and safe enough to be integrated into higher level tasks such as manipulation, vision, tele-operation, interaction with humans, which all require a strong capacity to face unforeseen events in an efficient way. A second objective is to develop and practice the team management policies and the software development methods that are necessary to develop such complex robotic applications and to uncover new scientific questions related to this unprecedented complexity. We have made significant progress in these two directions.
We have designed a fully automatic walking motion generation, which decides automatically when and where to make a step with respect to simple locomotion goals such as desired translation and rotation speeds. We have connected this walking motion generation directly with vision processes, orchestrating the locomotion of the robot directly from its vision of the environment, steering the robot by simply stating what is the desired view on its environment, facing for example the door or the table. We have also begun experimenting on the interaction with humans and proposed a controller based observer that allows the robot to decipher the locomotion of a human and even predict it one step ahead of time.
Concerning the realisation of complex tasks, our main result is that we have extended and generalised the classical algorithms introduced twenty years ago for the kinematic control of redundant robots to the case of prioritised control objectives involving both equalities and inequalities, where only equalities could be properly dealt with before. This enables dealing efficiently with obstacle avoidance and even virtual obstacle avoidance. We designed then a specific numerical solver for this problem, based on a hierarchic complete orthogonal decomposition of the constraint matrices, combined with an active-set method specifically designed for this hierarchic problem, which allows real-time implementations (a solution is obtained in a fraction of a millisecond on a standard CPU).
This is where research on humanoid robots becomes interesting. It is not clear today whether a humanoid shape might be of interest when designing a robot, and for which applications. But a clear interest of humanoid robots today is that they concentrate in one single research platform most of the scientific and technological challenges that emerge when considering 'complex interactions with uncontrolled environments'. And while appearing first as bare integrations of technological and scientific achievements developed formerly and independently, these platforms reach now a level of maturity that turns them into new driving forces for robotics research, bringing up whole new problems and challenges of unprecedented complexity and technical difficulty.
The main objective of this research and training project is to enhance the algorithms and control laws of existing humanoid robots in order to obtain a walking behaviour versatile and safe enough to be integrated into higher level tasks such as manipulation, vision, tele-operation, interaction with humans, which all require a strong capacity to face unforeseen events in an efficient way. A second objective is to develop and practice the team management policies and the software development methods that are necessary to develop such complex robotic applications and to uncover new scientific questions related to this unprecedented complexity. We have made significant progress in these two directions.
We have designed a fully automatic walking motion generation, which decides automatically when and where to make a step with respect to simple locomotion goals such as desired translation and rotation speeds. We have connected this walking motion generation directly with vision processes, orchestrating the locomotion of the robot directly from its vision of the environment, steering the robot by simply stating what is the desired view on its environment, facing for example the door or the table. We have also begun experimenting on the interaction with humans and proposed a controller based observer that allows the robot to decipher the locomotion of a human and even predict it one step ahead of time.
Concerning the realisation of complex tasks, our main result is that we have extended and generalised the classical algorithms introduced twenty years ago for the kinematic control of redundant robots to the case of prioritised control objectives involving both equalities and inequalities, where only equalities could be properly dealt with before. This enables dealing efficiently with obstacle avoidance and even virtual obstacle avoidance. We designed then a specific numerical solver for this problem, based on a hierarchic complete orthogonal decomposition of the constraint matrices, combined with an active-set method specifically designed for this hierarchic problem, which allows real-time implementations (a solution is obtained in a fraction of a millisecond on a standard CPU).