Evolution of visual controllers for obstacle avoidance in mobile robotics

Abstract

The purpose of this work is to automatically design vision algorithms for a mobile robot, adapted to its current visual context. In this paper we address the particular task of obstacle avoidance using monocular vision. Starting from a set of primitives composed of the different techniques found in the literature, we propose a generic structure to represent the algorithms, using standard resolution video sequences as an input, and velocity commands to control a wheel robot as an output. Grammar rules are then used to construct correct instances of algorithms, that are then evaluated using different protocols: evaluation of trajectories performed in a goal reaching task, or imitation of a hand-guided trajectory. A genetic program is applied to evolve populations of algorithms in order to optimize the performances of the controllers. The first results obtained in a simulated environment show that the evolution produces algorithms that can be easily interpreted and which are clearly adapted to the visual context. However, the resulting trajectories are often erratic, and the generalization capacities are poor. To improve the results, we propose to use a two-phase evolution combining imitation and goal reaching evaluations, and to add some constraints in the grammar rules to enforce a more generic behavior. The results obtained in simulation show that the evolved algorithms are more efficient and more generic. Finally, we apply the imitation based evolution on real sequences and test the evolved algorithms on a real robot. Though simplified by dropping the goal reaching constraint, the resulting algorithms behave well in a corridor centering task, and show certain generalization capacities.

Study of estimation of distribution algorithms applied to neuroevolution

“…ML has found success in applications such as data mining and difficult-to-program applications [30]. There are many approaches to ML and evolving Neural Networks was chosen because the approach is capable of performing tasks such as robotic control [2,22,33] 1 , vision [14,34], game playing [28,44], and classification [26]. This particular topic is of interest because there has been limited research in using an EDA for evolving Neural Networks.…”

“…Functions are nodes which take a number of inputs and provide some output. Terminals are the leaves of the tree and accept no input; they are usually variables, inputs, or values 2 . For example, for symbolic regression, a search for a formula that best fits the curve of input data, the functions would likely be any number of mathematical functions (multiply, divide, add, subtract, sine, cosine) and the terminals would be the input variables (X,Y, ...) or values (1, 2, 3, ...).…”

How Intelligent is your Intelligent Robot?

“…There are many robots with individual intelligence of different types; examples include reactive behaviours such as obstacle avoidance which has been designed by hand [3], or artificially evolved [1]. There are also many examples of robots capable of individual learning: one important class of learning algorithm allows a robot to simultaneously localise itself within its environment while building a map of that environment (SLAM) [28], another is reinforcement learning (RL), for a survey see [17].…”

Beyond black-box optimization: a review of selective pressures for evolutionary robotics

“…Three stages were used, from locating a large immobile target to tracking a smaller, moving one. Parker [151] followed a similar incremental strategy to evolve gaits for a hexapod robot; Barlow et al [8] did the same for controllers of simulated UAV, but using a MOEA instead of a mono-objective EA; Barate and Mazanera [7] employed two phases to evolve vision algorithms for mobile robots, where the first phase was based on behavior imitation and the second one on goal-reaching evaluations.…”