TransFusionOdom: Transformer-Based LiDAR-Inertial Fusion Odometry Estimation

Multimodal fusion of sensors is a commonly used approach to enhance the performance of odometry estimation, which is also a fundamental module for mobile robots. Recently, learning-based approaches garner the attention in this field, due to their robust nonhandcrafted designs. However, the question of How to perform fusion among different modalities in a supervised sensor fusion odometry estimation task? is one of the challenging issues still remaining. Some simple operations, such as elementwise summation and concatenation, are not capable of assigning adaptive attentional weights to incorporate different modalities efficiently, which makes it difficult to achieve competitive odometry results. Besides, the Transformer architecture has shown potential for multimodal fusion tasks, particularly in the domains of vision with language. In this work, we propose an end-to-end supervised Transformer-based LiDAR-Inertial fusion framework (namely TransFusionOdom) for odometry estimation. The multiattention fusion module demonstrates different fusion approaches for homogeneous and heterogeneous modalities to address the overfitting problem that can arise from blindly increasing the complexity of the model. Additionally, to interpret the learning process of the Transformer-based multimodal interactions, a general visualization approach is introduced to illustrate the interactions between modalities. Moreover, exhaustive ablation studies evaluate different multimodal fusion strategies to verify the performance of the proposed fusion strategy. A synthetic multimodal dataset is made public to validate the generalization ability of the proposed fusion strategy, which also works for other combinations of different modalities. The quantitative and qualitative odometry evaluations on the KITTI dataset verify that the proposed TransFusionOdom can achieve superior performance compared with other learning-based related works.