analiza aiqbench
This paper proposes a rotary position encoding that treats each token as a distribution over camera poses and sampling times rather than a single point, aimed at geometry-aware attention in robot policies. The reported gains are modest: about 1 point on 50 RoboTwin 2.0 simulation tasks, and a jump from 32.5% to 52.5% on two real-robot tasks, which is a small sample. Readers should weigh the 17.4% latency overhead over standard RoPE against those gains, and note that the speedups are measured against other encodings on one GPU type.
POD-RoPE: Rotary Position Encoding over Physical Observation Distributions for Embodied AI
arXiv:2610.33745v1 Announce Type: cross Abstract: Embodied perception requires positional representations that capture both physical geometry and the spatial and temporal extent of observations. We introduce POD-RoPE (Physical Observation Distribution RoPE), which represents token observation footprints as distributions over camera centers, viewing directions, and sampling times. Under independent sampling, expected rotations encode relative relationships between distributions; when each distribution collapses to a single address, the construction recovers pointwise RoPE. The same rotation statistics support Q/K matching and coherence-dependent V/O transformations, allowing observation geometry to guide both information selection and content aggregation. Per-token pre-aggregation avoids increasing sequence length or processing pairs of observation nodes. We evaluate POD-RoPE across four robot policy architectures, simulation and real-world manipulation, and complementary reconstruction, depth-estimation, and novel-view-synthesis tasks. Integrating POD-RoPE into Fast-WAM improves mean success from 91.83% to 92.88% across all 50 RoboTwin 2.0 tasks, and from 32.5% to 52.5% on two real-robot tasks on Sharpa North. Controlled reconstruction ablations support the contributions of distribution encoding and V/O transformations. In H20 model-block benchmarks, POD-RoPE achieves median forward-plus-backward speedups of $1.91\times$ over P-RoPE and $5.34\times$ over RayRoPE, with a median latency overhead of 17.4% over Standard RoPE. These results support physical observation distributions as an effective and computationally efficient basis for geometry-aware attention.
Abstract:Embodied perception requires positional representations that capture both physical geometry and the spatial and temporal extent of observations. We introduce POD-RoPE (Physical Observation Distribution RoPE), which represents token observation footprints as distributions over camera centers, viewing directions, and sampling times. Under independent sampling, expected rotations encode relative relationships between distributions; when each distribution collapses to a single address, the construction recovers pointwise RoPE. The same rotation statistics support Q/K matching and coherence-dependent V/O transformations, allowing observation geometry to guide both information selection and content aggregation. Per-token pre-aggregation avoids increasing sequence length or processing pairs of observation nodes. We evaluate POD-RoPE across four robot policy architectures, simulation and real-world manipulation, and complementary reconstruction, depth-estimation, and novel-view-synthesis tasks. Integrating POD-RoPE into Fast-WAM improves mean success from 91.83% to 92.88% across all 50 RoboTwin 2.0 tasks, and from 32.5% to 52.5% on two real-robot tasks on Sharpa North. Controlled reconstruction ablations support the contributions of distribution encoding and V/O transformations. In H20 model-block benchmarks, POD-RoPE achieves median forward-plus-backward speedups of $1.91\times$ over P-RoPE and $5.34\times$ over RayRoPE, with a median latency overhead of 17.4% over Standard RoPE. These results support physical observation distributions as an effective and computationally efficient basis for geometry-aware attention.
Robotics (cs.RO); Artificial Intelligence (cs.AI)
arXiv:2610.33745 [cs.RO]
(or arXiv:2610.33745v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2610.33745 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 296 cuvinte
Subiecte conexe
Bridging the Simulation-to-Experiment Gap with Generative Models using Adversarial Distribution Alignment
arXiv · 30 oct. 2026
ACTIVA: Amortized Causal Effect Estimation via Variational Autoencoders
arXiv · 23 oct. 2026
Reinforcement Learning under State and Outcome Uncertainty: A Foundational Distributional Perspective
arXiv · 22 sept. 2026
Low-rank Distributional Matrix Completion
arXiv · 5 iun. 2026