A curious machine for autonomous visual exploration
暂无分享,去创建一个
Passively accepting measurements of the world is not enough, as the data obtained is always incomplete, and the inferences made from it uncertain to a degree which is often unacceptable. Machines that operate autonomously will always be faced with this dilemma, and can only be successful if they play a much more active role. This paper presents such a machine. It deliberately seeks out those parts of the world which maximize the fidelity of its internal representations, and keeps searching until those representations are acceptable. This paradigm is called visual autonomous exploration, and the machine an autonomous explorer.
There are three major contributions in this thesis. The first is a theory of how to explore, or more precisely a framework for active inference. With it one can produce an engineering design that has the advantages of modularity, yet at the same time contains the computational methods needed to navigate the sensor to the best data. The navigational component can be implemented as feedback around classic “bottom-up” inference. It can be retro-fitted to existing passive systems and confer on them all the advantages of activity.
The second contribution is a set of mechanisms that provide a layer of task supervision and control over the navigational loop, and that ensure robust operation when the world does not meet the assumptions of the theory. Without these mechanisms the system operates reflectively. With them it becomes adaptable and exploratory.
The final contribution is a working implementation that can actively infer superellipsoid descriptions of articulated multi-part objects. The system is entirely bottom-up and does not depend on any a priori knowledge of the environment. To my knowledge it is the first to have successfully closed the loop between gaze planning and the inference of complex 3D models.