Event
New Task Representation for Dual-Arm Manipulation: Decoupled Encoder and Actions
1 reports · 1 sources
Franziska Herbert and co-authors published on arXiv a semantic-geometric graph task representation for dual-arm manipulation that jointly encodes object identity, inter-object semantic relations, and each object's motion history, consisting of an MPNN encoder and a Transformer decoder. The authors claim that the encoder processes only temporal scene graphs and produces a structured representation decoupled from action labels, so that only the decoder needs to be fine-tuned on small-scale robot datasets, allowing the encoder to be reused across different embodiments. The authors report that the method was tested on eleven dual-arm tasks across two datasets and achieved full task success on two real-robot dual-arm tasks. The above results are the claims of the paper's authors.
Report timeline
- Semantic-Geometric Task Representation for Bimanual Manipulation from Human Demonstrations to Robot Action Planning
The authors propose a semantic-geometric graph task representation that jointly encodes object identities, inter-object semantic relations, and each object's motion history, consisting of an MPNN encoder and a Transformer decoder; the encoder processes only the temporal scene graph and produces structured representations decoupled from action labels.
arXiv cs.RO ·