Workshop InformationThe video of our workshop has been avaliable Here.
Building a robot to navigate with human interaction is a topic that has been researched for a long time. Many methods are able to solve human interaction for robot navigation from diverse aspects. Computer vision methods like detection, segmentation and visual grounding enable the robot to understand the semantics in an environment. Imitation learning and reinforcement learning with a deep neural network contribute a lot for learning a robust navigation policy. Simultaneous localization and mapping (SLAM) and planning methods enable a robot for long-term navigation. On the other hand, embodied navigation following a natural language instruction attracts rising attention due to its wide application. Diverse tasks such as vision-language navigation (VLN), embodied question answering (EQA), visual-dialog navigation (VDN) are proposed.
There are many efforts to incorporate natural language processing and vision-language alignment into a navigation model to enable its language comprehension ability. Moreover, since there exists a large domain gap between a simulated environment and the real physical environment, some researchers adopt transfer learning methods to reduce the domain gap. Many research works have been devoted to related topics, leading to rapid growth of related publications in the top-tier conferences and journals.
This workshop will investigate current ways of human-robot interaction in the senario of robot navigation to give a promising direction of robot-human interaction. The topics of the workshop include (but not limited to):
- visual-based navigation
- Reinforcement learning, policy exploration for navigation
- Vision-based simultaneous localization and mapping (SLAM), planning
- Vision-language navigation, visual-dialog navigation and other cross-modal visual navigation
- Vision-language grounding, visual commonsense reasoning
- Navigation applications involving humans, e.g. indoor robots, UAV, auto-driving, gaming AI, etc.
- New datasets and metrics to evaluate the benefit of the robot navigation approaches for the specific embodied navigation problems