Reluctance coil guns are electromagnetic launchers having a good ratio of energy transmitted to actuator volume, making them a good choice for propelling objects with a limited actuator space. In this paper, we focus on an application, which is launching real size soccer balls with a size constrained robot. As the size of the actuator cannot be increased, kicking strength can only be improved by enhancing electrical to mechanical energy conversion, compared to existing systems. For this, we propose to modify its inner structure, splitting the coil and the energy storage capacitor into several ones, and triggering the coils successively for propagating the magnetic force in order to improve efficiency. This article first presents a model of reluctance electromagnetic coil guns using a coupled electromagnetic, electrical and mechanical models. Four different coil gun structures are then simulated, concluding that splitting the kicking coil into two half size ones is the best trade-off for optimizing energy transfer, while maintaining an acceptable system complexity and controllability. This optimization results in robust enhancement and leads to an increase by 104 % of the energy conversion compared to a reference launcher used. This result has been validated experimentally on our RoboCup robots. This paper also proves that splitting the coil into a higher number of coils is not an interesting trade-off. Beyond results on the chosen case study, this paper presents an optimization technique based on mixed mechanic, electric and electromagnetic modelling that can be applied to any reluctance coil gun.

No abstract

This paper presents a smart embedded Functional Electrical Stimulator (FES), able to stimulate a muscle only when a specific movement pattern occurs. This pattern is detected using an inertial measurement unit (IMU) coupled with a feature detector and a neural classifier. Architecture of the FES is first presented, then embedded processing algorithms composed of feature extraction and neural network classification are detailed. Results show that the muscle vibration happening when stimulation is needed can be recognized in more than 90% of cases using less than 3% of average embedded processor resources on a ARM M4F.