← Back ICRA 2026

Privacy-Aware LLMs-Assisted Task Planning for Home Robots

Zhanjie Chen, Weihua Sheng

PDF

AI summary

Key figure (auto-extracted from paper)

A two-stage cloud-edge architecture using a local LLM for privacy filtering enables home robots to safely leverage cloud LLMs for task planning without compromising user privacy or task performance.

Privacy-Preserving Robotics Cloud-Edge Architecture LLM Task Planning Home Service Robots Semantic Privacy Filtering Edge AI

Problem

Cloud-based LLMs enhance robotic task planning but require uploading sensitive visual data to third-party servers, creating critical privacy risks in home environments. Existing anonymization techniques often distort task-relevant information or lack the contextual reasoning needed for nuanced privacy protection.

Approach

The system uses a lightweight local LLM to screen images, extract semantic descriptions, and judge privacy relevance before transmitting only masked images and necessary text to a cloud LLM for high-level planning.

Key results

Novel two-stage cloud-edge hybrid architecture for privacy-aware robotic planning
Local LLM filter accurately distinguishes low-discriminability private objects from non-private ones
End-to-end system preserves high task success rates while minimizing cloud data leakage
Outperforms general-purpose baselines in privacy judgment without modifying cloud models

Why it matters

Enables safe, socially acceptable deployment of intelligent home robots by balancing robust cloud-based reasoning with strict on-device privacy protection.

Abstract

Multi-modal large language models (LLMs) are expected to significantly enhance the intelligence of home service robots. However, reliance on cloud processing of raw visual data poses critical privacy risks. To address this problem, we propose a novel two-stage cloud-edge hybrid architecture for robots in domestic environments. This architecture employs a lightweight local LLM to perform sensitive content screening and semantic abstraction before transmitting the data to a more powerful cloud-based LLM for high-level planning and reason- ing. Experiments with our end-to-end system demonstrate that it effectively protects a wide range of private data with minimal impact on task success rates. Without modifying cloud models, our approach offers a deployable performance–privacy trade- off for home robots, advancing safe and socially acceptable autonomy.

Index terms

Task Planning Service Robotics