Dex-Net as a Service (DNaaS)

Pusong Li, Bill DeRose, Jeffrey Mahler, Ajay Kumar Tanwani, Ken Goldberg (UC Berkeley, AUTOLAB) and Juan Aparicio Ojea (Siemens)

  • Notice

    ** Currently in beta release. Please report bugs and suggestions to Pusong Li at **
  • Description

    Anyone can upload a 3D object mesh (in .obj format with triangular faces) and visualize candidate grasps for a parallel-jaw gripper, each ranked by their robustness to inherent uncertainty/errors in sensing, control, and physics (green for most robust).

    Computing robust grasps currently requires up to 3 minutes; this version is single-threaded and does not support simultaneous users.

    Dex-Net 1.0 estimates the probability of force closure under sampled uncertainty in object and gripper pose and friction. Dex-Net 2.0 and 3.0 include Deep Learning using the the Dex-Net 1.0 sampling model to synthesize training data.


    Dex-Net as a Service (DNaaS): A Cloud-Based Robust Robot Grasp Planning System. Pusong Li, Bill DeRose, Jeffrey Mahler, Juan Aparicio Ojea, Ajay Kumar Tanwani, Ken Goldberg. IEEE International Conference on Automation Science and Engineering (CASE), Munich, Germany, August 2018.

    Dex-Net 1.0: A Cloud-Based Network of 3D Objects for Robust Grasp Planning Using a Multi-Armed Bandit Model with Correlated Rewards. Jeffrey Mahler, Florian T. Pokorny, Brian Hou, Melrose Roderick, Michael Laskey, Mathieu Aubry, Kai Kohlhoff, Torsten Kroeger, James Kuffner, Ken Goldberg. IEEE International Conference on Robotics and Automation, (ICRA), Stockholm, Sweden. May 2016. Finalist for Best Manipulation Paper Award.


    The DNaaS project is sponsored in part by NSF, NDSEG, Siemens, Google, Amazon Robotics, Toyota Research Institute, Autodesk, ABB, Samsung, Knapp, Loccioni, Honda, Intel, Comcast, Cisco, Hewlett-Packard, PhotoNeo, and NVidia.

    Additional Information

    For more information, papers, and videos, please see the Dex-Net project on Github, or read our FAQs.