Reliability-Aware Temporal Memory for Long-Term Drift Reduction in End-to-End Visual Odometry under Domain Shift
Abstract
Abstract Visual Odometry (VO) is essential for mobile robot navigation in GPS-denied environments. Learningbased VO reduces dependence on hand-engineered pipelines by training the model directly on data. Under domain transition, these systems suffer from drift in long routes. This paper proposes a reliability-aware temporal memory framework for long-term drift alleviation in end-to-end VO. The framework stores visual-temporal representations only when they are both reliable and novel. A heteroscedastic uncertainty head estimates input-dependent pose uncertainty and converts it into a frame-level reliability score. During retrieval, stored entries are weighted using feature similarity and stored-frame reliability. The retrieved memory context is then combined with the current representation through adaptive convex fusion. The framework is evaluated in a domain-controlled indoor simulation environment, with supporting validation on standard indoor benchmarks (TUM RGB-D and ICL-NUIM). The evaluation assesses trajectory accuracy, long-term drift, domain-shift robustness, reliability-aware memory behavior, ablation performance, and computational efficiency. The proposed method achieves an absolute trajectory error (ATE) of 0.089m, drift of 1.2% per meter, and 38 FPS in the reported implementation. The ATE increases to 0.201m under combined domain shift, compared with 0.512m for the CNN–Transformer baseline.
Read the paper
Related studies
- Diarylethenes for Memories and Switches
- Long-Term Potentiation and Memory
- Social Memory Studies: From “Collective Memory” to the Historical Sociology of Mnemonic Practices
- The Cognitive Neuroscience of Working Memory
- Memory aging and brain maintenance
- Effector and Memory CTL Differentiation
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.