Research on Reliable and Safe Occupancy Grid Prediction in Underground Parking Lots

TL;DR

This work tackles the gap in indoor autonomous driving research by focusing on occupancy grid prediction in underground parking lots. It combines CARLA-based data collection with the SurroundOcc occupancy network to generate dense 3D occupancy ground truth and train multi-camera occupancy predictions, addressing challenges unique to indoor environments. The study demonstrates that training on parking-lot data improves IoU-based occupancy predictions, while identifying limitations in ground-truth generation and semantic labeling that warrant further real-world data and methodological refinements. Overall, the approach advances perception robustness and safety for autonomous vehicles operating in confined indoor spaces, with implications for scalable indoor deployment and future planning integration.

Abstract

Against the backdrop of advancing science and technology, autonomous vehicle technology has emerged as a focal point of intense scrutiny within the academic community. Nevertheless, the challenge persists in guaranteeing the safety and reliability of this technology when navigating intricate scenarios. While a substantial portion of autonomous driving research is dedicated to testing in open-air environments, such as urban roads and highways, where the myriad variables at play are meticulously examined, enclosed indoor spaces like underground parking lots have, to a significant extent, been overlooked in the scholarly discourse. This discrepancy highlights a gap in derstanding the unique challenges these confined settings pose for autonomous navigation systems. This study tackles indoor autonomous driving, particularly in overlooked spaces like underground parking lots. Using CARLA's simulation platform, a realistic parking model is created for data gathering. An occupancy grid network then processes this data to predict vehicle paths and obstacles, enhancing the system's perception in complex indoor environments. Ultimately, this strategy improves safety in autonomous parking operations. The paper meticulously evaluates the model's predictive capabilities, validating its efficacy in the context of underground parking. Our findings confirm that the proposed strategy successfully enhances autonomous vehicle performance in these complex indoor settings. It equips autonomous systems with improved adaptation to underground lots, reinforcing safety measures and dependability. This work paves the way for future advancements and applications by addressing the research shortfall concerning indoor parking environments, serving as a pivotal reference point.

Figures (18)

• Figure 1: This figure shows the timeline of this work. First, we gathered extensive data reflective of real-world parking dynamics based on the creation of a meticulous simulated underground parking lot environment within CARLA. The collected data is then used to generate intensive occupancy labels. Next, leveraging these meticulously crafted occupancy labels, we embarked on a supervised learning process for the "SurroundOcc" model. Finally, the prediction results are evaluated and summarized
• Figure 2: Data collection flow chart. We can obtain the map information needed for collection, the number of collection scenes, and vehicle configuration from the configuration file. Then, the procedure will generate a world to start collecting data. In this process, the procedure will save data in Nuscenes data set format and process relational data to save the relational database. In particular, the lidarseg file is extracted from the lidar file in the samples folder.
• Figure 3: The pipeline of the proposed method. Initially, data collection is undertaken within a simulated parking lot environment. This process entails gathering point cloud data, which is subsequently utilized to produce detailed occupancy labels. Concomitantly, the camera footage encapsulated within the amassed dataset serves as the input for the SurroundOcc model, enabling it to perform occupancy prediction tasks, thereby harnessing the rich visual information for an enhanced understanding of the parking space. The generated dense occupancy labels supervise this process.
• Figure 4: The sensor installation location of the nuScenes official acquisition vehicle caesar2020nuscenes. Six cameras and five millimeter-wave radars are mounted in different directions to collect circumnavigation data. Lidar is mounted on the top of the car to scan the surrounding environment.
• Figure 5: This is an image collected by camera. A frame of data collected during vehicle driving is displayed