Semantic Scene Representations

Organisers

Christian Rauch*, Linus Nwankwo*, Shail Jadav#, Jun Zhang+

* Chair of Cyber Physical Systems, TU Leoben
# Autonomous Systems Lab, TU Wien
+ Institute of Visual Computing, TU Graz

Workshop Description

Autonomous systems that operate in the real world and around humans have special requirements on their generalisability and explainability. The increasing complexity of these systems and the tasks that we expect them to take over from us in future calls for novel, high-level interfaces to these systems grounded in semantic concepts from natural language. We expect that language-grounded representations facilitate scene understanding, generalisation to unseen environments and tasks, the explainability of an autonomous agent’s actions to a human, and we expect that these natural language interfaces will increase the accessibility of such complex systems for the general public without the need for domain-specific knowledge.

The workshop presents and discusses recent advances in semantic and natural-language-based 3D scene and environment representations, localisation and mapping techniques, as a building block for a variety of interactive robotic tasks.

Topics:

  • Open-set vision-language models for robotics
  • Human-robot natural language interfaces
  • Planning and reasoning on abstract task goals
  • Object detection and state estimation
  • Learning multi-modal representations
  • Explicit and implicit 3D scene representations
  • Interactive robot tasks (manipulation, navigation, ...)
  • Semantic and neural-based SLAM
  • Benchmarks and metrics for open-set object identification and localisation
  • Perception for safe human robot interaction
  • Language-guided robot motion policies
  • Visual localisation with neuromorphic sensor (event camera)
  • Generalizable neural rendering and Gaussian Splatting
  • Multi-view stereo and structure from motion

Paper Submission

We accept full papers (8 pages) and short abstracts (2-4 pages) of already published work. Please refer to the general Call for Papers for details on the expected submission format and the AIRoV template.

You can submit your contribution directly to our workshop (Track: "SSR") via CMT.

Programme

Time Paper Title Speaker
TBD TBD TBD