English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Robotic grasping for a diverse set of objects is essential in many robotmanipulation tasks. One promising approach is to learn deep grasping modelsfrom large training datasets of object images and grasp labels. However,empirical grasping datasets are typically sparsely labeled (i.e., a smallnumber of successful grasp labels in each image). The data sparsity issue canlead to insufficient supervision and false-negative labels and thus results inpoor learning results. This paper proposes a Maximum Likelihood Grasp SamplingLoss (MLGSL) to tackle the data sparsity issue. The proposed method supposesthat successful grasps are stochastically sampled from the predicted graspdistribution and maximizes the observing likelihood. MLGSL is utilized fortraining a fully convolutional network that generates thousands of graspssimultaneously. Training results suggest that models based on MLGSL can learnto grasp with datasets composing of 2 labels per image. Compared to previousworks, which require training datasets of 16 labels per image, MLGSL is 8x moredata-efficient. Meanwhile, physical robot experiments demonstrate an equivalentperformance at a 90.7% grasp success rate on household objects.

Learn to Grasp with Less Supervision: A Data-Efficient Maximum Likelihood Grasp Sampling Loss