Jul 2018 – Apr 2022 · 3 yrs 9 mos · Oakland, California, United States

• Project Lead of the Generalized Behaviors Team
• Developing a synthetic data pipeline for training 3D perception networks
• Managing a fleet of robots that are currently deployed in homes across California
• Implemented a variety of manipulation behaviors (I.e. place/grasps/open) for home tidying

Sep 2013 – Jul 2018 · 4 yrs 10 mos · Berkeley, CA

• Research Experience
• Machine Learning: Imitation Learning , Multi-Armed Bandits, Policy Gradients
• Computer Vision: Object Detection, CNNs, GANs
• Robotics: Motion Planning, Grasp Planning
• Software: EC2, Amazon Mechanical Turk, Tensorflow, PyTorch, ROS, Ray, OpenCV
• Managerial: Mentored up to Ten Bachelors and Master Students, Experience in Recruiting and Interviewing
• Here is a sample of recent projects:
• Algorithmic Development for Robust Imitation Learning:
• Developed DART a new algorithm for Imitation Learning that is competitive with state of the art methods such as DAgger. However unlike DAgger, DART is computationally inexpensive, user-friendly and is safe. A blog post on this can be found at http://bair.berkeley.edu/blog/2017/10/26/dart/
• Simulation for Self-Driving Cars:
• The goal is to create fast python based simulations, which can enable rapid prototyping of algorithms for self-driving car agents. Our system first uses Object Detection to extracted driving behavior from live traffic camera streams. Then applies DART to train multi-agent deep policies over a distributed EC2 network.
• Robotic Manipulation for the Home :
• Leveraging recent advances in Imitation Learning and Object Detection, such as Faster R-CNNs and YOLO, we have trained a Toyota HSR robot to perform various home tasks, such as picking up playing cards or retrieving objects from a cupboard. Recently, we have applied DART to train the HSR robot to perform the sequential task of making a bed. A video can be found at https://youtu.be/HbNNkrk9aOI
• Order Fulfillment in Warehousing:
• Implemented an end-to-end prototype system for packing boxes using the Dex-Net Grasping Software and an Industrial ABB YuMi Robot. Given a heap of objects the robot would search for a given set and pack them into a box. A video of the system can be found at https://www.youtube.com/watch?v=i6K3GI2_EgU
• See here for more information and complete list of publications:
• https://people.eecs.berkeley.edu/~laskeymd/

May 2013 – Aug 2014 · 1 yr 3 mos

• Created a SLAM based algorithm using machine learning with an IMU and a pedometer. Applied algorithm to allow users to walk around a virtual world with a tablet.