# On-Device Robotic Planning: Eliminating Inference Redundancy for Efficient Decision-Making

Source: [On-Device Robotic Planning: Eliminating Inference Redundancy for Efficient Decision-Making](<https://www.sota2.com/research/paper/6a1d1238dbd7150506f841aa>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-02T05:00:12.522Z

Authors: Joonhee Lee, Hyunseung Shin, Hyunmi Kim, Pei Zhang, Jeonggil Ko

Published: 2026-05-29T15:51:42.000Z

## Abstract

Reasoning-based robotic policies using large language and vision-language models achieve strong semantic planning capabilities but mostly suffer from a high inference latency that limits practical real-time deployment. In this work, we observe that robotic reasoning workloads contain substantial temporal redundancy, where consecutive observations frequently produce identical actions and subgoals. Based on this insight, we present REIS, a human cognition inspired robotic decision-making framework that minimizes unnecessary reasoning while preserving semantic adaptability. REIS combines lightweight scene gating, KV-steered affordance routing, and deliberative reasoning to accelerate robotic control under embodied constraints. Experiments on ALFRED, and real-world robotic tasks demonstrate that REIS significantly suppresses reasoning overhead while maintaining competitive task performance.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Action Learning on ALFRED (Q3)](<https://www.sota2.com/research/sota/action-learning-on-alfred-q3.md>): Action Learning; ALFRED (Q3); Accuracy: 54.8.

- [Action Learning on ALFRED (Q1)](<https://www.sota2.com/research/sota/action-learning-on-alfred-q1.md>): Action Learning; ALFRED (Q1); Accuracy: 92.5.

- [Manipulation on Realworld Manipulation Pick and Place](<https://www.sota2.com/research/sota/manipulation-on-realworld-manipulation-pick-and-place.md>): Manipulation; Realworld Manipulation Pick and Place; Episode Accuracy: 72.9.

- [Action Learning on ALFRED (Q2)](<https://www.sota2.com/research/sota/action-learning-on-alfred-q2.md>): Action Learning; ALFRED (Q2); Accuracy: 89.4.

- [Navigation on Realworld Navigation](<https://www.sota2.com/research/sota/navigation-on-realworld-navigation.md>): Navigation; Realworld Navigation; Episode Accuracy: 64.1.

- [Robot Instruction Following on ALFRED](<https://www.sota2.com/research/sota/robot-instruction-following-on-alfred.md>): Robot Instruction Following; ALFRED; Episode Accuracy: 51.2.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2605.31460>)

[PDF](<https://arxiv.org/pdf/2605.31460.pdf>)