The problem we address in this paper is how to plan and execute high quality paths for robots subject to nonholonomic constraints while navigating obstacles in 2D space. The navigation is to be carried out continuously at speed and may be subject to drift that is not predictable a priori. The problem raises the challenge of adaptively maintaining a smooth robust path of low computational cost. The algorithm is complete in providing feasible paths connecting to the goal in cluttered environments without global maps or positioning while also optimising the path curvature in free space. The approach is a generic gradient-based methodology set in dynamic potential fields that are not subject to fixed local minima or other misdirecting surface features of static fields. Multiple planning and execution cycles are interleaved to allow frequent updates for dealing with unanticipated obstacles and drift. We present our methodology and demonstrate experimental results for simulated robots. The results show that low curvature paths are found that robustly connect to the goal under perturbation through a sequence of fast adaptive replanning.

This paper addresses the problem of how to enable smoothness generically in nonholonomic motion planning for mobile robots travelling at speed both without and with unpredictable drift. The problem raises the challenge of maintaining goal connection while also optimising the path curvature. The approach is a gradient based methodology set in dynamic potential fields and has the capacity to enable integration with existing potential field based methods for obstacle avoidance and coordinated motion. The methodology interleaves multiple planning and execution cycles to deal with drift. We present our methodology and demonstrate experimental results for the 4-D unicycle and a nonholonomic integrator. The results show that the curvature metric used is able to improve smoothness to a given degree of optimality while reliably maintaining goal connection. They also show that interleaving and increased smoothness allows the path to keep relatively close to optimised plans.

It is well known that performing gradient descent method on fixed surfaces may result in poor travel through getting stuck in local minima and other surface features. Subgoal chaining in supervised learning is a method to improve travel for neural networks by directing local variation in the surface during training. This paper shows that linear subgoal chains such as those used in ERA are not sufficient to overcome the local minimum problem and examines nonlinear subgoal chains as a possible alternative.

This paper considers the performance of cross-validation across runs in terms of efficiency and accuracy and a method for improving it. A heuristic method loosely inspired by Mitchell's concept and version spaces technique is proposed for recognising when and to what extent the learning runs obtain an optimal generalisation performance. The approach used, the neural bidirectional convergence (NBDC), converges towards a solution from dual pairs of directions. The pair members provide complementary information for each other that is unavailable to uni-directional learning and which allows candidate concept elimination. Tests are carried out on classification problems in comparison with standard uni-directional cross-validation. The results indicate that NBDC is able to terminate learning at either more efficient junctures or with better generalisation accuracy or both depending on the problem.