Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ACORN: Adaptive Contrastive Optimization for Safe and Robust Fine-Grained Robotic Manipulation

Zhongquan Zhou, Shuhao Li, Zixian Yue

TL;DR

ACORN targets safety and robustness in fine-grained robotic manipulation under distribution shifts by introducing a safety-centric metric ecosystem and a plug-and-play adaptive contrastive optimization. It combines dual-perturbation contrastive sampling with a curriculum-informed objective that jointly optimizes robust regression and trajectory discrimination, while focusing on high-priority joints to boost safety without sacrificing performance. Across diverse manipulation tasks, ACORN yields substantial gains in safety metrics (notably up to 23% improvement in ACR-F under moderate noise) and demonstrates energy efficiency benefits, with ablations confirming the value of curriculum and margin settings. The approach is compatible with existing ACT architectures and offers a practical path toward reliable real-world deployment and potential sim-to-real transfer.

Abstract

Embodied AI research has traditionally emphasized performance metrics such as success rate and cumulative reward, overlooking critical robustness and safety considerations that emerge during real-world deployment. In actual environments, agents continuously encounter unpredicted situations and distribution shifts, causing seemingly reliable policies to experience catastrophic failures, particularly in manipulation tasks. To address this gap, we introduce four novel safety-centric metrics that quantify an agent's resilience to environmental perturbations. Building on these metrics, we present Adaptive Contrastive Optimization for Robust Manipulation (ACORN), a plug-and-play algorithm that enhances policy robustness without sacrificing performance. ACORN leverages contrastive learning to simultaneously align trajectories with expert demonstrations while diverging from potentially unsafe behaviors. Our approach efficiently generates informative negative samples through structured Gaussian noise injection, employing a double perturbation technique that maintains sample diversity while minimizing computational overhead. Comprehensive experiments across diverse manipulation environments validate ACORN's effectiveness, yielding improvements of up to 23% in safety metrics under disturbance compared to baseline methods. These findings underscore ACORN's significant potential for enabling reliable deployment of embodied agents in safety-critical real-world applications.

ACORN: Adaptive Contrastive Optimization for Safe and Robust Fine-Grained Robotic Manipulation

TL;DR

ACORN targets safety and robustness in fine-grained robotic manipulation under distribution shifts by introducing a safety-centric metric ecosystem and a plug-and-play adaptive contrastive optimization. It combines dual-perturbation contrastive sampling with a curriculum-informed objective that jointly optimizes robust regression and trajectory discrimination, while focusing on high-priority joints to boost safety without sacrificing performance. Across diverse manipulation tasks, ACORN yields substantial gains in safety metrics (notably up to 23% improvement in ACR-F under moderate noise) and demonstrates energy efficiency benefits, with ablations confirming the value of curriculum and margin settings. The approach is compatible with existing ACT architectures and offers a practical path toward reliable real-world deployment and potential sim-to-real transfer.

Abstract

Embodied AI research has traditionally emphasized performance metrics such as success rate and cumulative reward, overlooking critical robustness and safety considerations that emerge during real-world deployment. In actual environments, agents continuously encounter unpredicted situations and distribution shifts, causing seemingly reliable policies to experience catastrophic failures, particularly in manipulation tasks. To address this gap, we introduce four novel safety-centric metrics that quantify an agent's resilience to environmental perturbations. Building on these metrics, we present Adaptive Contrastive Optimization for Robust Manipulation (ACORN), a plug-and-play algorithm that enhances policy robustness without sacrificing performance. ACORN leverages contrastive learning to simultaneously align trajectories with expert demonstrations while diverging from potentially unsafe behaviors. Our approach efficiently generates informative negative samples through structured Gaussian noise injection, employing a double perturbation technique that maintains sample diversity while minimizing computational overhead. Comprehensive experiments across diverse manipulation environments validate ACORN's effectiveness, yielding improvements of up to 23% in safety metrics under disturbance compared to baseline methods. These findings underscore ACORN's significant potential for enabling reliable deployment of embodied agents in safety-critical real-world applications.
Paper Structure (23 sections, 9 equations, 4 figures, 3 tables)

This paper contains 23 sections, 9 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Safety in Robot Learning
  4. Imitation Learning
  5. Distribution Shift
  6. Preliminaries
  7. ACT Algorithm
  8. Contrastive Learning
  9. Methodology
  10. Safety-Centric Evaluation Metrics
  11. Noise Resilience Evaluation
  12. Contrastive Sample Generation via Dual Perturbation
  13. ACT-ACORN Objective Function
  14. Huber Loss
  15. Contrastive Loss
  16. ...and 8 more sections

Figures (4)

  • Figure 1: ACORN Overview: A three-stage optimization strategy for safety-aware reinforcement learning.
  • Figure 2: The contrastive loss contribution increases exponentially as baseline loss $\mathcal{L}_b$ decreases below unity.
  • Figure 3: High-priority Joint Group: (Left) ACT vs. (Right) ACT-ACORN
  • Figure 4: Secondary Joint Group: (Left) ACT vs. (Right) ACT-ACORN