Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AutoLoop: Fast Visual SLAM Fine-tuning through Agentic Curriculum Learning

Assaf Lahiany, Oren Gal

TL;DR

AutoLoop tackles the challenge of adding loop-closure robustness to learning-based visual SLAM without excessive training cost. It combines an offline precomputed loop-closure database with a DDPG-driven curriculum that automatically adjusts the loop-closure weight during fine-tuning on a DPVO-based backbone, yielding comparable or superior loop-closure handling while reducing training steps by about an order of magnitude. The key innovations are a NetVLAD-SIFT offline loop-detection pipeline, a loop-closure aware loss, and an adaptive weight policy that minimizes loop error without degrading pose accuracy. The approach achieves practical improvements in outdoor robotics scenarios, preserves real-time inference, and reduces data and compute requirements for adapting SLAM systems to new environments, though indoor generalization remains a challenge and future work could explore online adaptation and broader architecture applicability.

Abstract

Current visual SLAM systems face significant challenges in balancing computational efficiency with robust loop closure handling. Traditional approaches require careful manual tuning and incur substantial computational overhead, while learning-based methods either lack explicit loop closure capabilities or implement them through computationally expensive methods. We present AutoLoop, a novel approach that combines automated curriculum learning with efficient fine-tuning for visual SLAM systems. Our method employs a DDPG (Deep Deterministic Policy Gradient) agent to dynamically adjust loop closure weights during training, eliminating the need for manual hyperparameter search while significantly reducing the required training steps. The approach pre-computes potential loop closure pairs offline and leverages them through an agent-guided curriculum, allowing the model to adapt efficiently to new scenarios. Experiments conducted on TartanAir for training and validated across multiple benchmarks including KITTI, EuRoC, ICL-NUIM and TUM RGB-D demonstrate that AutoLoop achieves comparable or superior performance while reducing training time by an order of magnitude compared to traditional approaches. AutoLoop provides a practical solution for rapid adaptation of visual SLAM systems, automating the weight tuning process that traditionally requires multiple manual iterations. Our results show that this automated curriculum strategy not only accelerates training but also maintains or improves the model's performance across diverse environmental conditions.

AutoLoop: Fast Visual SLAM Fine-tuning through Agentic Curriculum Learning

TL;DR

AutoLoop tackles the challenge of adding loop-closure robustness to learning-based visual SLAM without excessive training cost. It combines an offline precomputed loop-closure database with a DDPG-driven curriculum that automatically adjusts the loop-closure weight during fine-tuning on a DPVO-based backbone, yielding comparable or superior loop-closure handling while reducing training steps by about an order of magnitude. The key innovations are a NetVLAD-SIFT offline loop-detection pipeline, a loop-closure aware loss, and an adaptive weight policy that minimizes loop error without degrading pose accuracy. The approach achieves practical improvements in outdoor robotics scenarios, preserves real-time inference, and reduces data and compute requirements for adapting SLAM systems to new environments, though indoor generalization remains a challenge and future work could explore online adaptation and broader architecture applicability.

Abstract

Current visual SLAM systems face significant challenges in balancing computational efficiency with robust loop closure handling. Traditional approaches require careful manual tuning and incur substantial computational overhead, while learning-based methods either lack explicit loop closure capabilities or implement them through computationally expensive methods. We present AutoLoop, a novel approach that combines automated curriculum learning with efficient fine-tuning for visual SLAM systems. Our method employs a DDPG (Deep Deterministic Policy Gradient) agent to dynamically adjust loop closure weights during training, eliminating the need for manual hyperparameter search while significantly reducing the required training steps. The approach pre-computes potential loop closure pairs offline and leverages them through an agent-guided curriculum, allowing the model to adapt efficiently to new scenarios. Experiments conducted on TartanAir for training and validated across multiple benchmarks including KITTI, EuRoC, ICL-NUIM and TUM RGB-D demonstrate that AutoLoop achieves comparable or superior performance while reducing training time by an order of magnitude compared to traditional approaches. AutoLoop provides a practical solution for rapid adaptation of visual SLAM systems, automating the weight tuning process that traditionally requires multiple manual iterations. Our results show that this automated curriculum strategy not only accelerates training but also maintains or improves the model's performance across diverse environmental conditions.
Paper Structure (15 sections, 8 equations, 3 figures, 6 tables)

This paper contains 15 sections, 8 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Methodology
  4. Off-Line Loop Closure Database
  5. Baseline Model
  6. DPVO Loss Supervision
  7. Loop Closure Aware Loss
  8. Agentic Curriculum-Learning
  9. Loop Closure Aware Fine-Tuning
  10. Experiments
  11. Pre-compute Loop Closure Pairs
  12. Fine Tuning
  13. Benchmarks Comparison
  14. Computational Analysis
  15. Conclusion

Figures (3)

  • Figure 1: Loop Closure Detection Pipeline: The system processes TartanAir sequences through three stages to identify and validate loop closure pairs.
  • Figure 2: Pre-computed loop closure pairs distribution on TartanAir training set. A total of 337 scenes processed with 551 loop closure pairs detected.
  • Figure 3: Progression of the loop weight during DPVO fine-tuning using the incorporated loop closure loss variant.