Towards Robust Multi-Object Pose Estimation under Sparse and Noisy Data in Robotic Systems

초록

This paper addresses the problem of estimating the poses of multiple objects under sparse and noisy data in robotics systems in unknown environments. Due to the data sparsity caused by limited data acquisition policies for saving resources and the data noises caused by environmental disturbance, the robustness of the conventional methods for the estimation degenerates. This paper proposes a robust multiobject pose estimation method to solve the problem, which can be applicable to not only ground environments but also space environments. First, a robot builds a map as a model its surrounding environments by simultaneous localization and mapping (SLAM). Then, multiple objects are detected by a visual sensor with a deep learning-based algorithm. Finally, the poses of multiple objects are robustly estimated by probabilistic filters with range data and localize them in the built map. The performance of the proposed method was tested with a real robotic system with visual and range sensors. Experimental results showed that the proposed method can robustly estimate the poses of multiple objects under sparse and noisy data.