Projects

Motion Planning Framework

The idea of the MoPa framework is to provide a suitable C++ scenario and bindings for motion and path planning algorithms over static and dynamic environments. In simpler words, the MoPa framework is a 2D world with several objects that moves. Then, an intelligent agent can be added towards finding ... [More] a way to move from an initial point to a predefined goal, trying not to cruash, touch or being hit by an object. The framework also tries to define all the architectural interfaces to be able to interchange any module (for instance the graphic engine, or the world's physics) for new implementations. [Less]

egocentric

In this project, we attempted an approach to path planning based on our own innovative expandable circular egocentric map. The map, composed of neurons, is capable of varying its coverage through adjustment of a weight factor. Through activations of the neurons in the map, we find that our ... [More] autonomous agents can adequately perform crowd and obstacle avoidance. We also implemented attention control, which react naturally to both dynamic and static objects in the environment. 1.1 Source of Inspiration The role of artificial intelligence in gaming has seen an exponential rise over the last few years. With the scope of games growing larger and larger with every new release, it is not uncommon for gamers to find themselves immersed in large scale environments with hundreds and thousands of non playing characters. The importance of these, ‘not so important’ characters come to the fore in order for a gamer to lose him or herself in a game. Life-like steering motions of these characters thus become quintessential for a truly immersive gaming experience. This paper presents a novel technique that makes use of multi-layered Egocentric Spatial Maps for steering and path planning of autonomous agents. Under this implementation, agents can navigate themselves in dynamically changing environments performing collision avoidance around static as well as dynamic objects. For attention control, the implementation was based on Gazenet architecture. With this, each agent has a head which can respond accordingly to objects in the environment as well. The implementation of this technique is carried out under the OpenSteer framework which is an open source C++ library that aids in the construction of steering behaviors for autonomous characters in games and animation. [Less]