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

Sudden changes in the dynamics of robotic tasks, such as contact with anobject or the latching of a door, are often viewed as inconvenientdiscontinuities that make manipulation difficult. However, when thesetransitions are well-understood, they can be leveraged to reduce uncertainty oraid manipulation---for example, wiggling a screw to determine if it is fullyinserted or not. Current model-free reinforcement learning approaches requirelarge amounts of data to learn to leverage such dynamics, scale poorly asproblem complexity grows, and do not transfer well to significantly differentproblems. By contrast, hierarchical POMDP planning-based methods scale well viaplan decomposition, work well on novel problems, and directly consideruncertainty, but often rely on precise hand-specified models and taskdecompositions. To combine the advantages of these opposing paradigms, wepropose a new method, MICAH, which given unsegmented data of an object's motionunder applied actions, (1) detects changepoints in the object motion modelusing action-conditional inference, (2) estimates the individual local motionmodels with their parameters, and (3) converts them into a hybrid automatonthat is compatible with hierarchical POMDP planning. We show that modellearning under MICAH is more accurate and robust to noise than priorapproaches. Further, we combine MICAH with a hierarchical POMDP planner todemonstrate that the learned models are rich enough to be used for performingmanipulation tasks under uncertainty that require the objects to be used innovel ways not encountered during training.

Learning Hybrid Object Kinematics for Efficient Hierarchical Planning Under Uncertainty