Unknown Object

Research on unknown object handling focuses on enabling robots and autonomous systems to perceive, interact with, and learn from objects not previously encountered during training. Current efforts concentrate on developing robust perception models (e.g., neural networks, graph networks) and control algorithms (e.g., reinforcement learning, Bayesian optimization) that leverage various sensor modalities (vision, tactile, acoustic) for object reconstruction, pose estimation, and manipulation. This work is crucial for advancing robotics, autonomous navigation, and other applications requiring adaptable and generalizable perception in dynamic, open-world environments.