Multiple object trackingMultiple object tracking in point cloud sequences aims to detect objects of interest in each frame and associate them across time to construct coherent trajectories for each individual object. Existing methods typically adopt a tracking-by-detection frameworkTracking-by-detection framework, where tracking is performed based on object boxes generated by a pre-trained detector. However, this two-stage paradigm often leads to suboptimal performance, as it decomposes the problem into two separate stages: object detection and temporal associationTemporal association, rather than optimizing tracking in a unified or end-to-end mannerEnd-to-end manner. This chapter presents two novel joint detection and tracking methods: Sect. 7.3 introduces a 4D object tubelet representation that encodes the spatio-temporal locations of objects, and formulates the tracking problem as a tubelet detection task. Section 7.4 proposes a point-based tracking framework, which achieves end-to-end object tracking by introducing a tracking headTracking head on top of a point-based detection model. Extensive experiments on several public datasets demonstrate that both methods achieve promising performance while maintaining satisfactory efficiency.