Integrating Historical Learning and Multi-View Attention with Hierarchical Feature Fusion for Robotic Manipulation

Gaoxiong Lu, Zeyu Yan, Jianing Luo, Wei Li, Junzhi Yu · 2024

Read Integrating Historical Learning and Multi-View Attention with Hierarchical Feature Fusion for Robotic Manipulation by Gaoxiong Lu, Zeyu Yan, Jianing Luo, Wei Li, Junzhi Yu (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-11591612.txt.

Description

Gaoxiong Lu and co-authors improve robot manipulation with memory and multi-view attention in a 2024 paper. Robots usually act from the current view alone. That habit misses change over time and fails when the view is poor. Humans, by contrast, lean on past experience and shifting focus.

Their method blends past observations with past actions over time. A spatio-temporal attention step weighs what matters from history. A second module picks the most useful camera views automatically. Semantic and texture features are merged, like global and local human vision. Tests ran on a standard benchmark and in real-world scenes. The method handled varied tasks with strong robustness. Adaptability across tasks stood out in the results. The authors frame history as a missing sense in robot control. Fine-grained tasks benefit most from the fused features.

Catalog record

CREATORGaoxiong Lu, Zeyu Yan, Jianing Luo, Wei Li, Junzhi Yu
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC11591612

Browse the full library · About the Tower of Babel