1474 Part G Human-Centered and Life-Like Robotics the robot imitated him/her by pushing the car sideways. rehearsal of the various actions (akin to the aforemen- Starting from this simple experiment,we need to for- tioned theory of motor perception)was used to generate malize what is required for a system that has to acquire hypotheses to be compared with the actual sensory in- and deliver imitation. put.It is then remarkable how more recently a modified Roboticists have been fascinated by the discovery approach of this paradigm has been used in compari- of mirror neurons and the purported link to imitation son with real human transcranial magnetic stimulation that exists in the human nervous system.The litera-(TMS)data. ture on the topic extends from models of the monkey's Ito et al.[62.138](not the Masao Ito of cerebellar (nonimitative)action recognition system (see Oztop fame)took a dynamical systems approach using a re- and Arbib [62.112])to models of the putative role current neural network with parametric bias(RNNPB) of the mirror system in imitation (see Demiris and to teach a humanoid robot to manipulate certain objects Jolnson [62.119]and Arbib et al.[62.1291),and in In this approach the parametric bias(PB)encodes (tags) real and virtual robots (see Schaal et al.[62.130]).Oz-certain sensorimotor trajectories.Once learning is com- top et al.[62.131]propose a taxonomy of the models plete the neural network can be used either to recall of the mirror system for recognition and imitation,a given trajectory by setting the PB externally or pro- and it is interesting to note how different the com-vide input for the sensory data only and observe the PB putational approaches that have been now framed as vector that would represent in that case the recognition mirror system models are,including recurrent neural of the situation on the basis of the sensory input only(no networks with parametric bias(see Tani et al.[62.132)),motor information available).It is relatively easy to in- behavior-based modular networks (see Demiris and terpret these two situations as the motor generation and Johnson [62.119]),associative memory-based methods the observation in a mirror neurons model. (see Kuniyoshi et al.[62.1331),and the use of multiple The problem of building useful mappings be- direct-inverse models as in the MOSAIC architecture tween dissimilar bodies (consider a human imitating (Wolpert et al.[62.134];cf.the multiple paired forward-a bird's flapping wings)was tackled by Nehaniv and inverse models of Sect.62.3.2). Dautenhahn [62.136]where an algebraic framework for Following the work of Schaal et al.[62.130]and imitation is described and the correspondence problem Oztop et al.[62.131]we can propose a set of schemas formally addressed.Any system implementing imitation required to produce imitation: should clearly provide a mapping between either dissim- ilar bodies or even in the case of similar bodies when determine what to imitate,inferring the goal of the either the kinematics or dynamics is different depending demonstrator. on the context of the imitative action. establish a metric for imitation (correspondence;see Sauser and Billard [62.139]modeled the ideomotor Nehaniv [62.135]), principle,according to which observing the behavior of map between dissimilar bodies(mapping) others influences our own performances.The ideomotor imitate behavior formation, principle points directly to one of the core issues of the mirror system,that is,the fact that watching somebody which are also discussed in greater detail by Nehaniv and else's actions changes something in the activation of the Dautenhahn [62.136].In practice,computational and observer,thus facilitating certain neural pathways.The Part robotic implementations have tackled these problems work in question also gives a model implemented in with different approaches and emphasizing different terms of neural fields (see Sauser and Billard [62.139] parts or specific subproblems of the whole,for exam- for details)and tries to explain the imitative cortical ple,in the work of Demiris and Hayes [62.137],the pathways and the behavior formation. n 62.5 Extroduction As the foregoing makes clear,robotics has much to tional neuroethology helps us understand how the learn from neuroscience and much to teach neuro- brain of a creature has evolved to serve action- science.Neurorobotics can learn from the ways in oriented perception,and the attendant processes which the brains and bodies of different creatures of learning,memory, planning,and social inter- adapt to diverse ecological niches -as computa- action.1474 Part G Human-Centered and Life-Like Robotics the robot imitated him/her by pushing the car sideways. Starting from this simple experiment, we need to for￾malize what is required for a system that has to acquire and deliver imitation. Roboticists have been fascinated by the discovery of mirror neurons and the purported link to imitation that exists in the human nervous system. The litera￾ture on the topic extends from models of the monkey’s (nonimitative) action recognition system (see Oztop and Arbib [62.112]) to models of the putative role of the mirror system in imitation (see Demiris and Johnson [62.119] and Arbib et al. [62.129]), and in real and virtual robots (see Schaal et al. [62.130]). Oz￾top et al. [62.131] propose a taxonomy of the models of the mirror system for recognition and imitation, and it is interesting to note how different the com￾putational approaches that have been now framed as mirror system models are, including recurrent neural networks with parametric bias (see Tani et al. [62.132]), behavior-based modular networks (see Demiris and Johnson [62.119]), associative memory-based methods (see Kuniyoshi et al. [62.133]), and the use of multiple direct-inverse models as in the MOSAIC architecture (Wolpert et al. [62.134]; cf. the multiple paired forward￾inverse models of Sect. 62.3.2). Following the work of Schaal et al. [62.130] and Oztop et al. [62.131] we can propose a set of schemas required to produce imitation: • determine what to imitate, inferring the goal of the demonstrator, • establish a metric for imitation (correspondence; see Nehaniv [62.135]), • map between dissimilar bodies (mapping), • imitate behavior formation, which are also discussed in greater detail byNehaniv and Dautenhahn [62.136]. In practice, computational and robotic implementations have tackled these problems with different approaches and emphasizing different parts or specific subproblems of the whole, for exam￾ple, in the work of Demiris and Hayes [62.137], the rehearsal of the various actions (akin to the aforemen￾tioned theory of motor perception) was used to generate hypotheses to be compared with the actual sensory in￾put. It is then remarkable how more recently a modified approach of this paradigm has been used in compari￾son with real human transcranial magnetic stimulation (TMS) data. Ito et al. [62.138] (not the Masao Ito of cerebellar fame) took a dynamical systems approach using a re￾current neural network with parametric bias (RNNPB) to teach a humanoid robot to manipulate certain objects. In this approach the parametric bias (PB) encodes (tags) certain sensorimotor trajectories. Once learning is com￾plete the neural network can be used either to recall a given trajectory by setting the PB externally or pro￾vide input for the sensory data only and observe the PB vector that would represent in that case the recognition of the situation on the basis of the sensory input only (no motor information available). It is relatively easy to in￾terpret these two situations as the motor generation and the observation in a mirror neurons model. The problem of building useful mappings be￾tween dissimilar bodies (consider a human imitating a bird’s flapping wings) was tackled by Nehaniv and Dautenhahn [62.136] where an algebraic framework for imitation is described and the correspondence problem formally addressed. Any system implementing imitation should clearly provide a mapping between either dissim￾ilar bodies or even in the case of similar bodies when either the kinematics or dynamics is different depending on the context of the imitative action. Sauser and Billard [62.139] modeled the ideomotor principle, according to which observing the behavior of others influences our own performances. The ideomotor principle points directly to one of the core issues of the mirror system, that is, the fact that watching somebody else’s actions changes something in the activation of the observer, thus facilitating certain neural pathways. The work in question also gives a model implemented in terms of neural fields (see Sauser and Billard [62.139] for details) and tries to explain the imitative cortical pathways and the behavior formation. 62.5 Extroduction As the foregoing makes clear, robotics has much to learn from neuroscience and much to teach neuro￾science. Neurorobotics can learn from the ways in which the brains and bodies of different creatures adapt to diverse ecological niches – as computa￾tional neuroethology helps us understand how the brain of a creature has evolved to serve action￾oriented perception, and the attendant processes of learning, memory, planning, and social inter￾action. Part G 62.5