Tianyi Ko

Takuya Ikeda, Sergey Zakharov, Tianyi Ko et al. · gatech

This paper addresses the challenging problem of category-level pose estimation. Current state-of-the-art methods for this task face challenges when dealing with symmetric objects and when attempting to generalize to new environments solely through synthetic data training. In this work, we address these challenges by proposing a probabilistic model that relies on diffusion to estimate dense canonical maps crucial for recovering partial object shapes as well as establishing correspondences essential for pose estimation. Furthermore, we introduce critical components to enhance performance by leveraging the strength of the diffusion models with multi-modal input representations. We demonstrate the effectiveness of our method by testing it on a range of real datasets. Despite being trained solely on our generated synthetic data, our approach achieves state-of-the-art performance and unprecedented generalization qualities, outperforming baselines, even those specifically trained on the target domain.

Tianyi Ko

In this paper, we propose a single-actuator gripper that can lift thin objects lying on a flat surface, in addition to the ability as a standard parallel gripper. The key is a crawler on the fingertip, which is underactuated together with other finger joints and switched with a passive and spring-loaded mechanism. While the idea of crawling finger is not a new one, this paper contributes to realize the crawling without additional motor. The gripper can passively change the mode from the parallel approach mode to the pull-in mode, then finally to the power grasp mode, according to the grasping state. To optimize the highly underactuated system, we take a combination of black-box optimization and physics simulation of the whole grasp process. We show that this simulation-based approach can effectively consider the precontact motion, in-hand manipulation, power grasp stability, and even failure mode, which is difficult for the static-equilibrium-analysis-based approaches. In the last part of the paper, we demonstrate that a prototype gripper with the proposed structure and design parameters optimized under the proposed process successfully power-grasped a thin sheet, a softcover book, and a cylinder lying on a flat surface.