In this thesis we tackle the problem of goal-oriented adaptation of a robot hitting motion. We propose the parameters that must be learned in order to use and adapt a basic hitting motion to play minigolf. Then, two different statistical methods are used to learn these parameters. The two methods are evaluated and compared. To validate the proposed approach, a minigolf control module is developed for a robotic arm. Using the different learning techniques, we show that a robot can learn the non-trivial task of deciding how the ball should be hit for a given position on a minigolf field. The result is a robust minigolf-playing system that outperforms most human players using only a small set of training examples.
Looking for publications? All our latest publications are listed here, and you can also use our search functions to help find what you are looking for. Power users might also want to consider searching on the EPFL Infoscience site which provides advanced publication search capabilities.
We derive the theoretical performance of three bio-inspired odor source localization algorithms (casting, surge-spiral and surge-cast) in laminar wind flow. Based on the geometry of the trajectories and the wind direction sensor error, we calculate the distribution of the distance overhead and the mean success rate using Bayes inference. Our approach is related to particle filtering and produces smooth output distributions. The results are compared to existing real-robot and simulation results, and a good match is observed.
We report on real-robot odor source localization experiments carried out in an environment with obstacles in the odor plume. The robot was equipped with an ethanol sensor and a wind direction sensor, and the experiments were carried out in a wind tunnel, i.e. in a controlled environment. An enhanced version of the surge-spiral algorithm was used, which was augmented with a dedicate behavior to manage obstacles (avoid them, or follow their contour). We compare the results in terms of distance overhead and success rate, and discuss the impact of obstacles on plume traversal.
We introduce a novel bio-inspired odor source localization algorithm (surge- cast) for environments with a main wind ﬂow and compare it to two well-known algorithms. With all three algorithms, systematic experiments with real robots are carried out in a wind tunnel under laminar ﬂow conditions. The algorithms are compared in terms of distance overhead when tracking the plume up to the source, but a variety of other experimental results and some theoretical considerations are provided as well. We conclude that the surge-cast algorithm yields signiﬁcantly better performance than the casting algorithm, and slightly better performance than the surge-spiral algorithm.
We compare two well-known algorithms for locating odor sources in environments with a main wind flow. Their plume tracking performance is tested through systematic experiments with real robots in a wind tunnel under laminar flow condition. We present the system setup and show the wind and odor profiles. The results are then compared in terms of time and distance to reach the source, as well as speed in upwind direction. We conclude that the spiral- surge algorithm yields significantly better results than the casting algorithm, and discuss possible rationales behind this performance difference.