Acceleration Control for Dynamic Manipulation of a Robot Turning Over Objects
This study deals with dynamic object manipulation by a robot using a tool. In order to keep the contact between the held spatula and the manipulated object, a variety of movements satisfying conditions in acceleration dimension should be planned. However, it is quite difficult to assure the acceleration to stay in a certain range in case disturbances exist. Therefore, this study proposes a control architecture for acceleration control and discusses how to develop a controller suitable for dynamic object manipulation. Comparison through simulation and experimental results reveals the features of the proposed control system.
As many previous studies have shown, kinodynamic planning deals with constraints on acceleration. Although most of the studies have discussed on the way to plan the optimized trajectory, it is also important to discuss the control architecture assuring the command acceleration determined based on the optimized trajectory. PD control or PID control are typical and fundamental solutions for achieving a given trajectory, while acceleration responses may deviate from the constraint condition due to external factors such as disturbance and control delay. Disturbance observer (DOB) is one of the most well-known architecture of acceleration feedback control . Acceleration feedforward control is also well studied in other areas such as precision control . White et al. have compared these two methods using the same magnetic disc drive and concluded that DOB offers better results in a low-frequency range, while acceleration feedforward reacts faster . However, there are very few examples combining these two  because they deal with similar disturbances. It is guessed that the combination may simply end up with larger vibration due to their conflict. Therefore, this paper discusses a method combining DOB and acceleration feedforward. The solution for avoiding the conflict is to add acceleration command to the DOB as feedforward input, instead of adding measured acceleration. It stands for only feedforwarding inertia force effect of the planned trajectory and such a controller will cover a wider frequency range of the planned trajectory. This study evaluates the acceleration control performance in simulation and clarify the property in view of acceleration tracking performance. Experimental results of movements turning over a pancake also show the performance of the proposed controller and further support the claim.
This section describes the control scheme for dynamic manipulation. First, MPC for dynamic manipulation is introduced based on the example of turning over pancakes. Second, model of contact between an object and a tool is described for deriving the acceleration limit. Then, the control architecture effective in dynamic manipulation is proposed in the last subsection. This section describes the mechanism of kinodynamic planning for turning over an object. A pancake is selected as a typical object to be turned over.
Turning over a pancake is a good candidate for evaluating performance of dynamic object manipulation, since it is a typical example of motion with kinodynamic constraints and it also requires a rapid response. Since acceleration is the key factor for kinodynamic constraints, this study discussed the property of an acceleration control architecture for MPC based kinodynamic planning. The results of simulations and experiments reveal the following properties of acceleration control architectures. _ Both DOB and acceleration FF have an effect to reduce the delay and they extend the control bandwidth of acceleration control. _ FF is effective in an environment with small distur bances, while a feedback control by DOB is more effective in real environment with larger unpredictable disturbances. _ Combination of DOB and FF is effective in case the command trajectory includes high frequency components. It was also confirmed that the proposed method results in phase lead in a certain frequency range because both control have the delay compensation property.
 K. Tahara, S. Arimoto, and M. Yoshida: “Dynamic object manipulation using a virtual frame by a triple soft-fingered robotic hand,” in Proc. IEEE Int. Conf. on Robotics and Automation (ICRA), 2010.
 Y. Bekiroglu, J. Laaksonen, J. A. Jorgensen, V. Kyrki, and D. Kragic: “Assessing grasp stability based on learning and haptic data,” IEEE Trans. on Robotics, vol. 27, no. 3, pp. 616–629, 2011.
 M. Li, Y. Bekiroglu, D. Kragic, and A. Billard: “Learning of grasp adaptation through experience and tactile sensing,” in Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS 2014), 2014.
 J. J. Kuffner, S. Kagami, K. Nishiwaki, M. Inaba, and H. Inoue: “Dynamically-stable motion planning for humanoid robots,” Autonomous Robots, Vol.12, No. 1, pp. 105–118, (2002).
 G. Goretkin, A. Perez, R. Platt, and G. Konidaris: “Optimal Sampling- Based Planning for Linear-Quadratic Kinodynamic Systems,” in Proc. IEEE Int. Conf. on Robotics and Automation (ICRA2013), pp. 2429– 2436, (2013).
 D. Webb and J. van den Berg: “Kinodynamic rrt*: Asymptotically optimal motion planning for robots with linear dynamics,” in Proc. IEEE Int. Conf. on Robotics and Automation (ICRA2013), pp. 5054– 5061, (2013).
 T. Kunz and M. Stilman: “Probabilistically complete kinodynamic planning for robot manipulators with acceleration limits,” in Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS2014), pp. 3713–3719, (2014).
 D. Nguyen-Tuong and J. Peters: “Model learning for robot control: a survey,” Cognitive processing, Vol. 12, No. 4, pp. 319–340, (2011).
 P. B. Wieber, “Trajectory free linear model predictive control for stable walking in the presence of strong perturbations,” in Proc. 2006 6th IEEE-RAS Int. Conf. on Humanoid Robots, pp. 137–142, (2006).
 H. Diedam, D. Dimitrov, P. B. Wieber, K. Mombaur, and M. Diehl: “Online walking gait generation with adaptive foot positioning through linear model predictive control” in Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, pp. 1121–1126, (2008)