Adaptive robotic handling of known light objects- ON-464

Desired discipline(s): Engineering - computer / electrical, Engineering, Engineering - mechanical, Computer science, Mathematical Sciences
Company: Gradient Robotics Inc
Project Length: 4 to 6 months
Preferred start date: 05/01/2021
Language requirement: English
Location(s): Toronto, ON, Canada; Canada
No. of positions: 2
Search across Mitacs’ international networks - check this box if you’d also like to receive profiles of researchers based outside of Canada: 
No

About the company: 

Gradient Robotics is a technology company that is shaping the future of online fulfillment. We develop and implement turnkey robotic-fulfillment solutions through advanced machine vision and artificial intelligence (AI). After having early traction in the online grocery space, our collaborative Autonomous Mobile Robots (AMR) are ready to enable online fulfillment that is fast, local and convenient. We have innovated the movement of materials, but still other repetitive and tedious tasks exist in a typical warehouse environment.

Please describe the project.: 

The project is about developing an effective computer vision system that can achieve human performance in dynamic pick and place for known retail products (cases and items of irregular shape, texture, sizes). The existing pick and place systems are rigid and space-inefficient. Therefore, not compatible with our space efficiency, collaborative work environment and form factor goals.

Final product will be a computer vision software that can communicate with a robotic arm (5-6 DOF manipulator) to tell ‘which item to pick next’, ‘where the item is’ and ‘how to pick’. There is also a hardware design portion since it goes hand-to-hand with the smart software, namely an end effector (contact surface with the items such as: gripper or suction). The robotic arm/manipulator will be outsourced/bought after successful demonstration in digital environment.

Main tasks to be performed by the candidate:

  • Researching state-of-the-art (literature scan);
  • Determining the best method for the job and the levels of tolerance, precision, accuracy that can be achieved using various existing state-of-art;
  • Preparing data and a test environment (could be simple error estimation or 2d/3d simulation)
  • Prototyping and benchmarking of basic item pick & place modules in a test environment
  • Iteration and Training: The measurable end goal is to achieve 95% accuracy and high precision with test datasets using reasonable compute

Methodology/techniques to be used:

  • Project management: Scrum/Agile methodology, Data Science Project Lifecycle, Discord and Email as communication media
  • Code management : Github (Git versioning)
  • Computer/Machine Vision (Camera and/or depth)
  • Data Preparation and Analysis (Data versioning)
  • Machine learning and Artificial neural networks

Required expertise/skills: 

  • Degree level: Masters
  • Specific software, specific skills. Optional: assets
  • Advanced software development knowledge: Python/C++
  • Computer Vision (OpenCV or equivalent)
  • Practical knowledge in Machine Learning and artificial neural networks
  • GPU-accelerated training (Tensorflow, PyTorch, CUDA or equivalent) 
  • PLUS : GPU-accelerated inference (CUDA, OpenCL or equivalent)
  • PLUS : Experience with robotics systems