Regularized Graph Matching for Correspondence Identification under Uncertainty in Collaborative Perception

Gao, Peng; Guo, Rui; Lu, Hongsheng; Zhang, Hao

doi:10.15607/rss.2020.xvi.012

Cited by 15 publications

(9 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The conventional methods require that the appearance and spatial pattern of objects must be unique, which are not robust to the perception uncertainty caused by occlusion, noisy data and model bias. Recently, regularized graph matching method is proposed [17], which addresses the observation uncertainty by adding regularization terms into the graph matching formulation. However, this method can not address the uncertainty in the graph matching model, and is not able to quantify the correspondence uncertainty caused by the perception uncertainty.…”

Section: A Correspondence Identificationmentioning

confidence: 99%

“…Minimizing this loss function during training is equivalent to maximizing the similarity of correct correspondences and minimizing the similarity of non-covisibile objects and matching uncertainty. During execution, given the quantified uncertainty in the identified correspondence, we further improve the correspondences results by defining a threshold λ, in order to remove the correspondences with high uncertainty values [17]. Specifically, if H(E(p(Y) i,i )) ≥ λ, the correspondence Y i,i is removed.…”

Section: Reducing Perceptual Non-covisibility and Correspondence Uncertaintymentioning

confidence: 99%

“…We construct each observation as a graph with node attributes generated from appearance features [17]. The edges are generated by Delaunay triangulation given the 2D camera coordinates of objects in SFAT and RFAT and 3D real world coordinates of objects in SMRC.…”

Section: A Experimental Setupsmentioning

confidence: 99%

“…• Multi-order graph matching (MOGM) [5], which integrates multiple different attributes in a learning-free way to identify correspondences. • Regularized graph matching (RGM) [17], which addresses perception uncertainty and non-covisible objects in a learning-free way to identify correspondences. • Graph convolutional network-based graph matching (GCN-GM) [11], which identifies correspondences by only optimizing the loss of overall similarity between two observations.…”

Section: A Experimental Setupsmentioning

confidence: 99%

“…Collaborative perception has been widely applied in a variety of real-world applications including human-robot collaborative assembly [18,20], multi-robot search and rescue [1,45], and connected autonomous driving [19,49]. Correspondence identification is defined as a problem to identify the same objects observed by multiple agents in their own fields of view, which is considered an essential component to enable collaborative perception [14,17,43]. For example, as illustrated by Figure 1, when a collaborative robot assists a human worker who wears an augmented reality (AR) headset to assemble a chair, they need to identify the correspondence of the chair parts in order to ensure that both the robot and the human correctly refer to the same object.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Bayesian Deep Graph Matching for Correspondence Identification in Collaborative Perception

Gao¹,

Zhang²

2021

Robotics: Science and Systems XVII

Self Cite

View full text Add to dashboard Cite

Correspondence identification is essential for multirobot collaborative perception, which aims to identify the same objects in order to ensure consistent references of the objects by a group of robots/agents in their own fields of view. Although recent deep learning methods have shown encouraging performance on correspondence identification, they suffer from two shortcomings, including the inability to address non-covisibility in collaborative perception that is caused by occlusion and limited fields of view of the agents, and the inability to quantify and reduce uncertainty to improve correspondence identification. To address both issues, we propose a novel uncertainty-aware deep graph matching method for correspondence identification in collaborative perception. Our new approach formulates correspondence identification as a deep graph matching problem, which identifies correspondences based upon graph representations that are constructed from the agents' observations. We introduce a novel deep graph matching network under the Bayesian framework to explicitly quantify uncertainty in the identified correspondences. In addition, we design a novel loss function that explicitly reduces correspondence uncertainty and perceptual non-covisibility during learning. We evaluate our approach in the robotics applications of collaborative assembly and multi-robot coordination using high-fidelity simulations and physical robots. Experiments have shown that, through addressing both uncertainty and non-covisibility, our approach achieves the state-of-the-art performance of correspondence identification.

show abstract

Section: A Correspondence Identificationmentioning

confidence: 99%

Section: Reducing Perceptual Non-covisibility and Correspondence Uncertaintymentioning

confidence: 99%

Section: A Experimental Setupsmentioning

confidence: 99%

Section: A Experimental Setupsmentioning

confidence: 99%