What does this research mean for the field?

Representing link-wise coefficient tensors of Bernstein polynomial Signed Distance Field (SDF) models in low-rank tensor formats (CP and TT) significantly reduces GPU runtime and memory usage while preserving sufficient geometric and gradient accuracy for real-time collision-aware robotics applications. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To develop a compact and differentiable low-rank tensor formulation for Bernstein polynomial signed distance field models to enable rapid GPU-based collision checking and motion control.

September 15, 2026Robotics and Autonomous SystemsOpen Access

Fast GPU evaluation of differentiable Signed Distance Fields for robotics via tensor decompositions

View Full Paper

Ask AI

Bookmark

Authors

AGAndrea Govoni SCSylvain Calinon GPGianluca Palli

Discussion

Member takes

Overview

Algorithmic analysis demonstrates faster evaluation and reduced memory for robot signed distance fields via tensor decompositions, highlighting efficient real-time collision avoidance on GPUs.

Key Points

To develop a compact and differentiable low-rank tensor formulation for Bernstein polynomial signed distance field models to enable rapid GPU-based collision checking and motion control.
Represented link-wise Bernstein polynomial coefficient tensors using low-rank canonical polyadic (CP) and tensor train (TT) tensor decomposition formats.
Conducted a roofline-inspired computational analysis to evaluate execution throughput, memory footprints, and runtime behaviors across polynomial orders and batch sizes.
Evaluated the differentiable models on robot manipulator link geometries within reactive collision-avoidance and trajectory generation tasks.
Low-rank tensor representations significantly decreased GPU runtime and memory footprints compared to dense models while preserving geometric accuracy for collision checking.
Canonical polyadic decomposition yielded the lowest nominal runtime in most configurations, whereas tensor train decomposition exhibited superior memory scalability.
Analytical gradient computations from compressed models provided accurate repulsion vectors for real-time robotic motion generation and reactive collision avoidance.

Cite This Study

Govoni et al. (2026) studied this question.

synapsesocial.com/papers/6aa9132f9013453be30a0efc https://doi.org/10.1016/j.robot.2026.105717

View Full Paper

Ask AI

Bookmark

September 15, 2026Robotics and Autonomous SystemsOpen Access

Fast GPU evaluation of differentiable Signed Distance Fields for robotics via tensor decompositions

View Full Paper

Ask AI

Bookmark

Authors

AGAndrea Govoni SCSylvain Calinon GPGianluca Palli

Discussion

Member takes

Overview

Algorithmic analysis demonstrates faster evaluation and reduced memory for robot signed distance fields via tensor decompositions, highlighting efficient real-time collision avoidance on GPUs.

Key Points

To develop a compact and differentiable low-rank tensor formulation for Bernstein polynomial signed distance field models to enable rapid GPU-based collision checking and motion control.
Represented link-wise Bernstein polynomial coefficient tensors using low-rank canonical polyadic (CP) and tensor train (TT) tensor decomposition formats.
Conducted a roofline-inspired computational analysis to evaluate execution throughput, memory footprints, and runtime behaviors across polynomial orders and batch sizes.
Evaluated the differentiable models on robot manipulator link geometries within reactive collision-avoidance and trajectory generation tasks.
Low-rank tensor representations significantly decreased GPU runtime and memory footprints compared to dense models while preserving geometric accuracy for collision checking.
Canonical polyadic decomposition yielded the lowest nominal runtime in most configurations, whereas tensor train decomposition exhibited superior memory scalability.
Analytical gradient computations from compressed models provided accurate repulsion vectors for real-time robotic motion generation and reactive collision avoidance.

Cite This Study

Govoni et al. (2026) studied this question.

synapsesocial.com/papers/6aa9132f9013453be30a0efc https://doi.org/10.1016/j.robot.2026.105717

View Full Paper

Ask AI

Bookmark