23 January 2018 in IEEE Robotics and Automation Letters
We consider a vehicle consisting of a robotic walking assistant pushed by a user. The robot can guide the person along a path and suggest a velocity by various means. The vehicle moves in a crowded environment and can detect other pedestrians in the surroundings. We propose a reactive planner that modifies the path in order to avoid pedestrians in the surroundings. The algorithm relies on a very accurate model to predict the motion of each pedestrian, i.e., the headed social force model. The possible trajectories for both the vehicle and the pedestrians are modeled as clothoid curves, which are efficient to manage from the numeric point of view and are very comfortable to follow for the user. Probabilistic techniques are used to account for the variability of the motion of each pedestrian. The path is efficient to generate, is collision free (up to a certain probability), and is comfortable to follow. Simulations and comparisons with a state-of-the-art planner using real data as well as experiments are reported to prove the effectiveness of the method.
24 December 2021 in IEEE Access
Generating accurate and efficient predictions for the motion of the humans present in the scene is key to the development of effective motion planning algorithms for robots moving in promiscuous areas, where wrong planning decisions could generate safety hazard or simply make the presence of the robot “socially” unacceptable. Our approach to predict human motion is based on a neural network of a peculiar kind. Contrary to conventional deep neural networks, our network embeds in its structure the popular Social Force Model, a dynamic equation describing the motion in physical terms. This choice allows us to concentrate the learning phase in the aspects which are really unknown (i.e., the model’s parameters) and to keep the structure of the network simple and manageable. As a result, we are able to obtain a good prediction accuracy even by using a small and synthetically generated training set. Importantly, the prediction accuracy remains acceptable even when the network is applied in scenarios radically different from those for which it was trained. Finally, the choices of the network are “explainable”, as they can be interpreted in physical terms. Comparative and experimental results prove the effectiveness of the proposed approach.
1 March 2022 in Robotics and Autonomous Systems
7 July 2020 in IEEE Robotics and Automation Letters
A combined minimum time - minimum jerk traffic management system for the vehicle coordination in an automated warehouse is presented. The algorithm is organised in two steps: in the first, a simple minimum time optimisation problem is solved, in the second step, this time-optimal solution is refined into a smooth minimum jerk plan for the autonomous forklifts in order to avoid impulsive forces that may unbalance the vehicle. For the first step, we propose a novel approach based on Linear Programming, which guarantees convergence to the optimal solution starting from a feasible point, and a low computational overhead, which makes it suitable for real-time applications. The output of this step is a piecewise constant velocity profile for all the moving robots that ensures collision avoidance. The second step takes such speed profile and generates its smoothed version, which minimises the jerk while respecting the same levels of safety of the solution generated by the first step. We discuss the different solutions with simulation and experimental data.