Path planning is crucial for Autonomous Mobile Robots applications. Traditionally, path planning based on human input and preferences has relied on hard to define reward-based learning or costly techniques requiring additional hardware. This work introduces a more accessible and flexible approach through sketch-guided imitation learning, where nontechnical users can simply draw the desired navigational path on a provided 2D map, which is then used to teach U-net models path planning behaviors. Additionally, the work draws on metrics from the fields of image generation and robotics to provide a novel evaluation framework. The approach is integrated into an end-to-end robotics stack to demonstrate its usability. The dataset and code are provided on https://github.com/charbel-a-hC/SKIPP.

International audience

Modular Multilevel Converters (MMCs) are widely used in medium to high voltage/power applications such as in energy transformation and grid connected systems. MMCs benefit from being redundant and allowing transformer-less operations. Unfortunately, they suffer from having a high cost due to the high number of sub-modules (SMs). Moreover, they are hard to control. In order to operate the MMCs properly, the load current, the circulating current and the capacitor voltages should be controlled. The usual control methods are synthesized in order to control the average lower and upper arm capacitor voltage as well as the load current. However, in some special cases, the capacitor values could change due to the aging of the capacitors, or the SM may stop working completely. Thus, creating an imbalance in the capacitor voltages of each arm in the system. This paper presents a new control scheme that allows to control the voltage of the upper and lower arms separately while controlling the load current at the same time. The usual and the new proposed controllers are developed and compared through software simulation. Finally, the conclusion is drawn.

Squirrel cage induction motors are widely used in the industrial field as they present many advantages compared to other types of electric machines. They are self-starting, need little maintenance, can work in harsh environment and are relatively cheap. Many industrial applications requiring current or speed control of the machine need the knowledge of the machine parameters. Therefore, induction motor parameter identification algorithms are developed taking in consideration the accuracy of the identification process which impacts on the quality of the control and thereby the overall system. These algorithms are usually sensitive to measurement noise thus there usage in noisy industrial environment is not desirable. This paper presents three induction motor offline identification algorithms and test their accuracy and their ability to work in noisy environment. The first method subject to the study is a Least Mean Square (LMS) based identification method working at standstill, the second is a newly introduced method robust toward noises and the third is the classic no-load and blocked rotor identification method. Simulation results are presented and finally conclusions are drawn.

This paper is dedicated to control-based energy optimization tools useful for the electric vehicles (EVs) propulsion chain, which puts in interaction an electric motor and its energy source static converter. It is essential in energy optimization studies to build an accurate simulation model that takes into account all possible losses of the propulsion chain components according to the operating point. For that reason, the present work provides an accurate loss analysis solution, by studying the various elements losses of an electric traction chain motorized by an Electrically-Excited Synchronous Motor (EESM). The machine power balance is established while taking copper and iron losses into account. Moreover, the inverter and chopper losses are evaluated while using a model based on a novel technique that consists in creating a modified version of output voltages affected by the inverter and chopper losses. Finally, all losses data are gathered in order to simulate the system on-line losses distribution and global efficiency for different operating conditions.

On-line system identification is generally used to find the mathematical or frequency model of a system for control applications. In particular, the cross-correlation technique is nowadays used for nonparametric identification of linear or linearized systems such as switching power converters. This method is however limited to systems with close time constants, therefore, it cannot be applied to systems with different time constants in a large bandwidth, such as machine drives. This paper presents a modified cross-correlation method. It first provides an improved calculation technique for the parameters of the pseudo-random binary signal used as input signal for identification. It then modifies the initial cross-correlation technique allowing the identification of systems with different time constants in a large bandwidth. Simulation results using MATLAB/Simulink validate the feasibility of the proposed technique in order to implement it using FPGAs.