OOD detection for 1D LiDAR scans
Out-of-Distribution (OOD) detection is a critical task in machine learning, particularly in autonomous systems where models are deployed in dynamic, unpredictable environments. A common architecture of OOD detection uses Variational Autoencoders (VAEs), which are generative models that are able to a...
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| مؤلفون آخرون: | |
| التنسيق: | Final Year Project |
| اللغة: | English |
| منشور في: |
Nanyang Technological University
2024
|
| الموضوعات: | |
| الوصول للمادة أونلاين: | https://hdl.handle.net/10356/181504 |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | Out-of-Distribution (OOD) detection is a critical task in machine learning, particularly in autonomous systems where models are deployed in dynamic, unpredictable environments. A common architecture of OOD detection uses Variational Autoencoders (VAEs), which are generative models that are able to able to capture complex data patterns, making them particularly useful in OOD detection. Autonomous Vehicles (AVs) utilise LiDAR scans amongst an array of other sensors for OOD detection, where recognizing anomalous data is critical for safety and reliability. However, VAEs have not been extensively evaluated for their accuracy in OOD detection specifically with LiDAR data, leaving a gap in the current research. This paper focuses on enhancing OOD detection in the context of LiDAR-based autonomous vehicle navigation, using VAEs and β-VAEs, the latter designed to encourage disentanglement in the latent space by adjusting a regularization parameter, β. Additionally, this study evaluates the models on varied track environments using the F1tenth framework, to measure of these OOD detection architectures across varied environments. |
|---|