This paper focuses on the quantization problem of connection weights of neural networks. Our previous work proposed a class of quantizers, called noise-shaping quantizer, for the quantization of neural networks. The performance of the proposed quantizer depends on the error diffusion filter. This paper proposes a systematic design method of error diffusion filters based on features of learning data, which is used for the learning of neural networks. In the proposed design method, satisfactory error diffusion filters are given by solving a kind of traveling salesman problems.

Human assist control of a control system driven by a human operator attracts much attention, and we have proposed a human assist control strategy has been proposed by using a control barrier function (CBF). However, the method is valid only for time-invariant environments; we cannot apply the method to a moving obstacle avoidance problem of a vehicle. In this paper, we introduce a time-invariant graph space as a time-varying state space. Then, we propose a time-varying CBF and human assist control based on the time-varying CBF for given time-varying environments. Finally, we confirm the effectiveness of the proposed method by computer simulation.

This paper presents a new whole-body path-following feedback control method for a multiple-steering snake-like robot based on Lyapunov stability theory. The locomotor has multiple links which are connected in a chain through revolute joints. Each link has a steering system at its midpoint. The locomotor transforms the periodical driving of its joints into its movement through the periodical operations of its steering systems. The control method enables to cause the endpoints of all the links to follow a desired path. This means that the whole-body motion of the locomotor can be specified by the shape of the path. All the wheels of the locomotor are passive, so that it has singular attitudes in which it cannot move in some directions. The locomotor is required to avoid singular attitudes while maintaining its desired motions, which means that the path must meander like a serpenoid curve representing a shape of a biological snake running. In the control method, the desired angular velocities of the joints for causing the tip of the top link to follow the path can be equalized to those for causing the ends of all the links to follow the path by control of the angles of all the steering systems. In other words, the locomotor has a particular structure which facilitates to achieve the desired motion of the top link and these of the other links, simultaneously. The asymptotical stability of the control method is guaranteed by Lyapunov's second method. Especially, the cost of the control input calculation is significantly reduced compared with that of another control method based on chained form which requires high order Lie derivatives. The validity of this control method is verified by computer simulations and by the comparison about the calculation time of the control inputs for the two methods.