PRANet: Point Cloud Registration with an Artificial Agent

TL;DR

PRANet introduces a novel approach to point cloud registration by framing it as a Markov Decision Process and training an artificial agent with deep supervised learning, achieving competitive results on standard datasets.

Contribution

It presents a new perspective by modeling registration as a sequential decision process and trains an agent end-to-end without reinforcement learning complexities.

Findings

Achieves results comparable or superior to state-of-the-art methods.

Effective on clean, noisy, and partially visible datasets.

Streamlined training process without experience replay buffer.

Abstract

Point cloud registration plays a critical role in a multitude of computer vision tasks, such as pose estimation and 3D localization. Recently, a plethora of deep learning methods were formulated that aim to tackle this problem. Most of these approaches find point or feature correspondences, from which the transformations are computed. We give a different perspective and frame the registration problem as a Markov Decision Process. Instead of directly searching for the transformation, the problem becomes one of finding a sequence of translation and rotation actions that is equivalent to this transformation. To this end, we propose an artificial agent trained end-to-end using deep supervised learning. In contrast to conventional reinforcement learning techniques, the observations are sampled i.i.d. and thus no experience replay buffer is required, resulting in a more streamlined training process. Experiments on ModelNet40 show results comparable or superior to the state of the art in the case of clean, noisy and partially visible datasets.