Binbin Xu,Allen Tao,Hugues Thomas,Jian Zhang,Timothy D. Barfoot
Abstract:In this paper, we introduce a LiDAR-based robot navigation system, based on novel object-aware affordance-based costmaps. Utilizing a 3D object detection network, our system identifies objects of interest in LiDAR keyframes, refines their 3D poses with the Iterative Closest Point (ICP) algorithm, and tracks them via Kalman filters and the Hungarian algorithm for data association. It then updates existing object poses with new associated detections and creates new object maps for unmatched detections. Using the maintained object-level mapping system, our system creates affordance-driven object costmaps for proactive collision avoidance in path planning. Additionally, we address the scarcity of indoor semantic LiDAR data by introducing an automated labeling technique. This method utilizes a CAD model database for accurate ground-truth annotations, encompassing bounding boxes, positions, orientations, and point-wise semantics of each object in LiDAR sequences. Our extensive evaluations, conducted in both simulated and real-world robot platforms, highlights the effectiveness of proactive object avoidance by using object affordance costmaps, enhancing robotic navigation safety and efficiency. The system can operate in real-time onboard and we intend to release our code and data for public use.
What problem does this paper attempt to address?
The paper attempts to address the problem of achieving proactive obstacle avoidance navigation for robots in dynamic indoor environments. Specifically, most existing robot navigation systems adopt reactive methods, which handle objects and humans based on simplified assumptions (e.g., constant speed). This may not be safe and efficient in complex and variable environments. The authors propose a new LiDAR-based object-level SLAM system that detects, tracks, and maps objects, and uses this information to generate an object-aware cost map for proactive obstacle avoidance. This approach aims to utilize the semantic information of objects to improve the safety and efficiency of navigation.
### Main Issues
1. **Limitations of Existing Navigation Systems**:
- Most existing navigation systems adopt reactive methods, handling objects and humans based on simplified assumptions (e.g., constant speed).
- This approach may not be safe and efficient in dynamic environments.
- Lack of sufficient utilization of object semantic information, leading to an inability to foresee potential movement risks of objects.
2. **Special Challenges of Indoor Environments**:
- The types and layouts of objects in indoor environments are more complex and variable.
- The presence of dynamic objects (e.g., moving furniture, pedestrians) increases the difficulty of navigation.
- Lack of sufficient indoor LiDAR datasets with semantic annotations limits the development of related research.
### Solution
1. **LiDAR-based Object-level SLAM System**:
- Use a 3D object detection network to identify objects in LiDAR keyframes.
- Refine the 3D poses of objects using the ICP algorithm.
- Perform data association and object tracking using a Kalman filter and the Hungarian algorithm.
- Update existing object poses and create new object maps for unmatched detections.
2. **Generating Object-aware Cost Maps**:
- Use the maintained object-level mapping system to create obstacle avoidance cost maps based on object semantics.
- The cost map considers the semantic attributes of different objects (e.g., the sliding risk of a chair), enabling proactive obstacle avoidance.
3. **Automatic Annotation Technique**:
- Introduce an automatic annotation technique that uses a CAD model database to generate accurate ground truth annotations, including object bounding boxes, positions, orientations, and point-level semantic information.
- Automatically annotate semantic points, instance points, object poses, and associate CAD models through a multi-session SLAM setup.
### Experimental Validation
- Extensive experiments were conducted in the Gazebo simulation environment and on real-world robot platforms to validate the system's effectiveness and real-time performance.
- Quantitative evaluation metrics (e.g., time to reach the goal, time spent in high-risk areas, travel distance) demonstrated the system's superiority.
- Compared to traditional conservative and opportunistic planners, the proposed system showed excellent performance in terms of safety and efficiency.
### Main Contributions
1. Propose a LiDAR-based object-level SLAM system capable of detecting, tracking, and mapping objects, and generating object-aware cost maps.
2. Introduce an automatic annotation technique to address the lack of semantic annotations in indoor LiDAR datasets.
3. Comprehensive validation through simulation and real-world experiments, demonstrating the impact of object-level semantic information on navigation performance.