Multi-robot plan adaptation by constrained minimal distortion feature mapping

Abstract

We propose a novel method for multi-robot plan adaptation which can be used for adapting existing spatial plans of robotic teams to new environments or imitating collaborative spatial teamwork of robots in novel situations. The algorithm selects correspondences between previous and current spatial features by the application of pairwise constraints, and generates the transformation function with a fast regular grid approximation which minimizes distortion. The algorithm requires minimal domain knowledge, is capable of transforming the spatial aspects of collaborative team behavior and performs better in noisy problems with large displacements than the most generally used quadratic differences method. The algorithm can be utilized for rapid plan adaptation, plan generalization or team behavior imitation. Methods are demonstrated on a multi-robot control problem in a random environment.

Efficient template-based path imitation by invariant feature mapping

“…In the field of both image processing and motion plan adaptation, research works have been done to morph a scene in the current context into a new scene by minimising the energy involved in morphing( [13], [14]). To address the "how-to" question, we can perceive a path imitation problem as plan adaptation whereby the original path(in our case, from demonstration) with the set of additional environmental features is projected to a warping space.…”

One-to-one feature matching with inaccurate maps

“…The main concept of maximizing average local consistency in RL is to seek a value for each individual q such that the two vectors are aligned thus yielding the maximum result. We start the conversion task by first disregarding the first constraint in (1) to allow q to have any value within range [0, 1]; following this scheme is the concept of continuous RL.…”

“…Finally, we wish to find values of the indicators such that they maximize objective (2) with respect to the one-to-one constraint x q(x) = 1 and properties (1). This problem is in the category of the quadratic assignment problem [19] and 0-1 quadratic programming where the complexity of solving for the optimal solution is NP-hard.…”

“…This is a sub-problem of navigation using inaccurate maps framework [1], [2] where the goal is to localize agents from environments into human-provided maps in order to retrieve proper actions. The human-provided maps are very convenient to create, for example from drawing, but they suffer from abstraction and from lacking of accuracy.…”

Towards One Shot Learning by imitation for humanoid robots

“…Based on the preservation of the spatial relationship between the imitated and the original features in that space, an imitated path can be generated. Such way of morphing a scene in the current context into a new scene exists in the field of both image processing and motion plan adaptation in which energy involved in morphing is minimised ( [12], [13]). …”