dGrasp: NeRF-Informed Implicit Grasp Policies with Supervised Optimization Slopes

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Sóti, Gergely, Huang, Xi, Wurll, Christian, Hein, Björn
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913561240928256
author Sóti, Gergely
Huang, Xi
Wurll, Christian
Hein, Björn
author_facet Sóti, Gergely
Huang, Xi
Wurll, Christian
Hein, Björn
contents We present dGrasp, an implicit grasp policy with an enhanced optimization landscape. This landscape is defined by a NeRF-informed grasp value function. The neural network representing this function is trained on simulated grasp demonstrations. During training, we use an auxiliary loss to guide not only the weight updates of this network but also the update how the slope of the optimization landscape changes. This loss is computed on the demonstrated grasp trajectory and the gradients of the landscape. With second order optimization, we incorporate valuable information from the trajectory as well as facilitate the optimization process of the implicit policy. Experiments demonstrate that employing this auxiliary loss improves policies' performance in simulation as well as their zero-shot transfer to the real-world.
format Preprint
id arxiv_https___arxiv_org_abs_2406_09939
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle dGrasp: NeRF-Informed Implicit Grasp Policies with Supervised Optimization Slopes
Sóti, Gergely
Huang, Xi
Wurll, Christian
Hein, Björn
Robotics
We present dGrasp, an implicit grasp policy with an enhanced optimization landscape. This landscape is defined by a NeRF-informed grasp value function. The neural network representing this function is trained on simulated grasp demonstrations. During training, we use an auxiliary loss to guide not only the weight updates of this network but also the update how the slope of the optimization landscape changes. This loss is computed on the demonstrated grasp trajectory and the gradients of the landscape. With second order optimization, we incorporate valuable information from the trajectory as well as facilitate the optimization process of the implicit policy. Experiments demonstrate that employing this auxiliary loss improves policies' performance in simulation as well as their zero-shot transfer to the real-world.
title dGrasp: NeRF-Informed Implicit Grasp Policies with Supervised Optimization Slopes
topic Robotics
url https://arxiv.org/abs/2406.09939