tailieunhanh - Adaptive Motion of Animals and Machines - Hiroshi Kimura et al (Eds) Part 9

Tham khảo tài liệu 'adaptive motion of animals and machines - hiroshi kimura et al (eds) part 9', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | Learning Energy-Efficient Walking with Ballistic Walking 159 The desired angle of the ankle joint is always fixed to 90 deg . Therefore the ankle joint works as a spring is attached. The simulation result of the controller is shown in Fig. 3 in which the resultant torque curves are shown with control mode during one period two steps . In this figure the control modes 1 2 3 and 4 correspond to swing I swing II swing III and support respectively. In Fig. 3 large torque is Fig. 3. State machine mode and torque during one period observed at the end of the swing phase and the beginning of the support phase. This torque might be caused by too large or too small torque applied at the beginning of the swing phase. If the appropriate torque is applied in swing I at the beginning of the swing phase this feedback torque might be lessen and the more energy-efficient walking could be realized. In the next section the optimization of this torque is attempted by adding a learning module. 3 Energy minimization by a learning module To realize the energy efficient walking a learning module which searches appropriate output torque in swing I is added to the controller described in the previous section . Besides torque the learning module searches the appropriate value of control parameters which determine the end of the duration of passive movement Tswg2. It is noted that these parameters are not related to the PD controller which stabilizes walking. For the evaluation of energy efficiency we use the average of all the torque which is applied during one walking period two steps Eval - - ip Tidt 10 T step J 0 i i Using this performance function the appropriate values of the parameters are searched in the probabilistic ascent algorithm as follows. 160 M. Ogino K. Hosoda M. Asada Fig. 4. Ballistic walking with learning module 1 2 if Eval Evalmin Amin A 3 Bmin B 4 Tswg2min Tswg2 5 A A random perturbation 6 B B random perturbation 7 Tswg2 Tswg2 random perturbation The simulation .