A Multimodal Perception-Driven Self Evolving Autonomous Ground Vehicle

Increasingly complex automated driving functions, specifically those associated with free space detection (FSD), are delegated to convolutional neural networks (CNNs). If the dataset used train network lacks diversity, modality, or sufficient quantities, driver policy that controls vehicle may induce safety risks. Although most autonomous ground vehicles (AGVs) perform well in structured surroundings, need for human intervention significantly rises when presented unstructured niche environments. To this end, we developed an AGV seamless indoor and outdoor navigation collect realistic multimodal data streams. We demonstrate one application of applied a self-evolving FSD framework leverages online active machine-learning (ML) paradigms sensor fusion. In essence, queries image against reliable stream, ultrasound, before fusing improve robustness. compare proposed prominent segmentation methods, DeepLabV3+ [1] . is state-of-the-art semantic model composed CNN autodecoder. consonance results, outperforms The performance attributed its ability self-learn space. This combination ML removes large datasets typically required by CNN. Moreover, technique provides case-specific classifications based on information gathered from scenario at hand.