Large Scale Deformable Object-Robot Interaction Simulation for Machine Learning in Robotics
||Large Scale Deformable Object-Robot Interaction Simulation for Machine Learning in Robotics|
||SNIC Medium Compute|
||Florian Pokorny <email@example.com>|
||Kungliga Tekniska högskolan|
||2021-02-01 – 2022-02-01|
||10207 10201 10299|
Deformable object manipulation is one of the key research frontiers in robotic manipulation research, with deformable objects ranging from rubber parts, ropes, cables, engine belts to every-day household items being currently beyond the ability of robots to interact with reliably.
In order to develop and test reliable machine learning methods to enable a robotic system to interact with such deformable materials, large training datasets are a key hurdle to advancements in this area. This compute resource application intends to utilize high-fidelity physics simulation to generate millions of simulations of deformable object - robot interaction sequences that will provide a first research dataset at this scale for the training of machine learning prediction models for deformable object manipulation.
This research will be conducted as part of a collaborative project between Chalmers University and KTH called DARMA: DAta-driven foundations for Robust deformable object MAnipulation with co-PIs Yiannis Karagiannidis (Chalmers) and Florian Pokorny (KTH) which also supports the PhD students that are listed as co-PIs for this compute resource application. DARMA is funded by the Wallenberg AI, Autonomous Systems and Software program: wasp-sweden.org