SHUFFLE TURN OF HUMANOID ROBOT SIMULATION BASED ON EMG MEASUREMENT

Transcription

1 SHUFFLE TURN OF HUMANOID ROBOT SIMULATION BASED ON EMG MEASUREMENT MASANAO KOEDA, TAKAYUKI SERIZAWA, AND YUTA MATSUI Osaka Electro-Communication University, Faculty of Information Science and Arts, Department of Computer Science Kiyotaki , Shijonawate, Osaka, , JAPAN. Recently, many researchers have been studying a method that involves the stepless slip motion of humanoid robots. However, it is not clear how it works or how it is controlled. In this research, we measured the muscle activity in the leg of a human subject performing a shuffle turning motion by using an electromyograph. The results indicated that the hip joint performed an important function in shuffle turning. Then, we verified this hypothesis using a life-size humanoid dynamic simulator that was constructed using ODE. By controlling the hip pitch joint angle and knee pitch angle using a proportional controller, the robot turned as much as a real human. Thus, our hypothesis was supported by this experiment. 1. Introduction Biped robots conventionally perform walking and turning motions through repeated foot stepping. However, foot stepping is inefficient, time consuming, unstable, and generally unsuitable in narrow spaces under constrained postures, as illustrated in Figure 1. There are great expectations with the use of biped robots in a kitchen, assembly line, and energy plant inspection/destruction applications. Recently, many researchers have been studying a method that involves the stepless slip motion of humanoid robots to realize smooth, quick, and high stability movement [1, 2, 3], as shown in Figure 1 Motivation 1

2 2 Figure 2. Miura et al. [4] proposed a control model for the slip turning motion of a humanoid robot. However, this model was not based on human motion and seems to be difficult to justify. In this research, we investigated the muscle activity in the leg of a real human subject performing shuffling motion by using an EMG (electromyograph). Based on this measurement data, we determined the joint motion of the leg required for the movement. Finally, we verified this result using a dynamic humanoid simulator, which was implemented using an accurately sized and weighted human model. Figure 2 Conceptual diagram Table 1 Measured leg muscles CH Muscle 1 Tensor fascia latae 2 Vastus laterails 3 Biceps femoris 4 Semitendinosus 5 Vastus medialis 6 Lateral head of gastrocnemius 7 Medial head of gastrocnemius 8 Tibialis anterior 2. Measurement of leg muscle activity during shuffle turning 2.1. Experimental methodology EMG sensors were attached to leg muscles to determine their activity during a shuffling motion. The attached positions of these EMG sensors are listed in Table 1 and Figure 3. In addition, the whole body motion was also simultaneously measured using a motion capturing system. The subject of this experiment was 22-year-old male, with 70 kg and a height of 165 cm (Figure 4). This subject performed 45 and 90 deg shuffle turning movements in upright standing and knee bending postures.

3 3 (a) Frontal view (b) Rear view Figure 3 EMG attached muscles Figure 4 Overview of subject with EMG P-EMGplus sensors (a wet-type sensor, with a maximum of 8 channels, 14-bit resolution, and 1-kHz sampling rate) manufactured by Oisaka Electronic Equipment Ltd. and a laptop PC (CF-W8, CPU: Intel Core2 Duo 1.2 GHz, RAM: 4 GB) manufactured by Panasonic were used for measuring the myoelectric potential of the leg. A VICON BLADE manufactured by Vicon was used for capturing the motion. This system is installed in Joint Institute for Advanced Multimedia Studio (JIAMS) of our university.

4 4 CH1 CH2 CH3 CH4 CH5 CH6 CH7 CH8 CH1 CH2 CH3 CH4 CH5 CH6 CH7 CH8 (a) Upright standing posture (b) Knee bending posture Figure 5 EMG of left leg during shuffle turning in 45 deg counterclockwise direction 2.2. Experiment and results The measured data shown by yellow lines on the EMG graph show the initial posture, start turning, and end turning moments. As you can see, CH1, CH2, CH7, and CH8 were active during turning (Figure 5). Taking into account these results, it is supposed that the hip joints are Table 2 Parameters of humanoid model in simulator rotated in the pitch axis direction and Name (ID No.) Size knee joints are of Body height (26) 1706[mm] Knee height (160) 429.6[mm] little relevance to Crotch height (158) 759.7[mm] shuffle turning. The Side neck height (31) 1440[mm] ankle joint should be Hip breadth (51) 340.3[mm] controlled Hip depth (167) 255.3[mm] horizontally to Thigh circumference (169) 536.4[mm] maintain the body s balance. Calf circumference (171) Foot length (174) Projected foot breadth (176) 366.8[mm] 249.4[mm] 96.3[mm] 3. Simulation Foot circumference (178) 247.8[mm] ODE (ver , double precision) was used to simulate the humanoid robot dynamics. The specifications of the computer were a CPU with Cervical height, sitting (76) Head height (3) Body weight (1) Trunk weight (-) Leg weight (-) 653.2[mm] 237.1[mm] 64.9[kg] 44.9[kg] 10.0[kg]

5 5 an Intel Core2 Duo 1.2-GHz processor and 4GB of RAM running the Windows7 32-bit OS. The simulated robot had 12 DOFs in the lower body, as illustrated in Figure 6. The foot size, link length of the leg, and weight of the body parts were determined by referring to [5] for a year-olds male Japanese. The detailed parameters are listed in Table 2. In this simulator, the dynamic friction coefficient was set to 0.2, empirically. Based on the result of the EMG measurement, to conduct the shuffle turning, only the hip pitch joint and ankle pitch joints were proportionally controlled in the simulation in both the upright standing and knee bending postures (Figure 7). The initial posture for the upright standing turning included 0 deg angles for the hip pitch, knee, and ankle pitch joints. To simulate {45, 90} deg shuffle turns, the target Figure 6 Simulation model angles for the hip pitch and ankle pitch joints were controlled in 0.2 deg steps in {100, 200} simulation loops, for total {increases, decreases} of {20, 40} deg, respectively, over approximately {3, 6} s. The final posture of the upright standing turning was the left hip at {20, 40} deg, right hip at {-20, -40} deg, knee at {0, 0} deg, left ankle at {-20, -40} deg, and right ankle at {20, 40} deg for the {45, 90} deg shuffle turns, respectively. Figure 7 Joint angle

6 6 (a) Upright standing posture (b) Knee bending posture Figure 8 45 deg shuffle turning in simulator In the same way, the initial posture of the knee bending turning was 30 deg for the hip pitch joint, 60 deg for the knee joint, and 30 deg for the ankle pitch joint. The same angle control was used in this simulation. The final posture of

7 7 the knee bending turning was the left hip at {50, 70} deg, right hip at {10, -10} deg, knee at {60, 60} deg, left ankle at {10, -10} deg, and right ankle at {50, 70} deg, respectively. As a result, as shown in Figure 8, the simulation worked much like real human motion, and our control method was validated experimentally. 4. Conclusion In this research, the EMG of the leg of a human during shuffling motion was measured, and it indicated that the hip joint motion plays an important function in shuffle turning. In addition, to verify this hypothesis, we built a life-size humanoid dynamic simulator using ODE, and simulated shuffle turning motions in two postures with very simple joint angle control. As a result, it was confirmed that this control method could turn the robot s body as much as the real motion of a human subject. In future work, we will specifically investigate whole body motion and examine the effect of the floor s friction on shuffle turning. References 1. M. Koeda, et al., Shuffle Turn and Translation of Humanoid Robots, In Proc. of 2011 IEEE Int. Conf. on Robotics and Automation, pp (2011). 2. K. Miura, et al., Analysis on a Friction Based Twirl for Biped Robots, In Proc. of 2010 IEEE Int. Conf. on Robotics and Automation, pp (2010). 3. K. Hashimoto, et al., Realization of Quick Turn of Biped Humanoid Robot by Using Slipping Motion with Both Feet, In Proc. of 2011 IEEE Int. Conf. on Robotics and Automation, pp , pp (2011). 4. K. Miura, et al., A Friction Based Twirl for Biped Robots, In Proc. of 8th IEEE-RAS Int. Conf. on Humanoid Robots, pp (2008). 5. Research Institute of Human Engineering for Quality Life, Japanese body size data (1997).