PointNet4D: A Lightweight 4D Point Cloud Video Backbone for Online and Offline Perception in Robotic Applications

TL;DR

PointNet4D is a lightweight, efficient 4D point cloud video backbone designed for real-time robotic perception, combining a hybrid Mamba-Transformer fusion and a novel pretraining strategy to improve performance across multiple tasks and datasets.

Contribution

It introduces PointNet4D, a novel lightweight backbone with a hybrid Mamba-Transformer fusion and a pretraining method, enabling efficient online and offline 4D perception for robotics.

Findings

Consistent performance improvements across 9 tasks and 7 datasets.

Effective in real-time robotic applications with resource constraints.

Substantial gains on RoboTwin and HandoverSim benchmarks.

Abstract

Understanding dynamic 4D environments-3D space evolving over time-is critical for robotic and interactive systems. These applications demand systems that can process streaming point cloud video in real-time, often under resource constraints, while also benefiting from past and present observations when available. However, current 4D backbone networks rely heavily on spatiotemporal convolutions and Transformers, which are often computationally intensive and poorly suited to real-time applications. We propose PointNet4D, a lightweight 4D backbone optimized for both online and offline settings. At its core is a Hybrid Mamba-Transformer temporal fusion block, which integrates the efficient state-space modeling of Mamba and the bidirectional modeling power of Transformers. This enables PointNet4D to handle variable-length online sequences efficiently across different deployment scenarios. To enhance temporal understanding, we introduce 4DMAP, a frame-wise masked auto-regressive pretraining strategy that captures motion cues across frames. Our extensive evaluations across 9 tasks on 7 datasets, demonstrating consistent improvements across diverse domains. We further demonstrate PointNet4D's utility by building two robotic application systems: 4D Diffusion Policy and 4D Imitation Learning, achieving substantial gains on the RoboTwin and HandoverSim benchmarks.