Our Research
We are primarily product focused and are driven by real world problems and challenges including:
- A factory environment
- Acquisition of perceptions
- Vision
- Store
- Printers
- Ink
- Projectors
- Scanners
- Point of Sale
- Robots
- Industries
- Support
We develop applications and products that will have a direct impact on the next generation of Epson robotic systems.
Our Innovation
Our work and research ideas include:
- Deep learning
- Reinforcement learning
- Transfer learning
- Virtual simulations
- Other machine learning disciplines
Academic Collaboration
To keep abreast with the latest theoretical advances in the field of robotics our R&D group is actively collaborating with academic partners on various projects.
Steven L. Waslander
Associate Professor, Institute for Aerospace Studies
Director, Toronto Robotics and AI Laboratory
In a factory environment, acquisition of perceptions on the operating environment is negatively impacted by many factors including lighting conditions, reflectance from object materials, occlusions, clutter scenes, limited field of view and limited resolution of the sensors.
The project explores the usage of actuated sensors to avoid aforementioned issues as an alternative solution to multi-sensor static clusters.
Dynamic sensor systems (DSSs), that is systems with actuated sensors, can be stabilized and directed at areas of interest independent of robot motion, leading to higher value and quality measurements.
Oliver Kroemer
Assistant Professor, Intelligent Autonomous Manipulation (IAM) Lab
Most industrial robot solutions in manipulation rely on specific programs designed by an expert. Those programs often follow fixed task plans and trajectories and therefore, rely on well-defined environments and exact object models. As a result, they limit the possible robot applications to repeatable tasks in structured environments.
The project focuses on usage of Deep Learning algorithm to create a closed loop-control and planning policy to generalize between a range of robot tasks, while capturing the detailed requirements of the individual tasks.
See: The Visual Cortex of Robotic System
The team seeks to improve on Image Sensing capabilities and functionalities to generate optimal image data sets and to automatically control and adapt to new environment lighting, object materials and scenes.
- Environment Control: Reinforcement Learning methods to generate policy for fast adaption to new factory environments and robotic tasks.
- Multi-modality Imaging: Deep Learning approaches to fuse info from different sensor types.
- Active Vision: Computer Vision & Reinforcement Learning methods to predict camera positions to minimize data acquisition cycle time while optimizing the quality of image data sets.
Think: The Cerebrum of Robotic System
The team leverages extensive experience in the fields of Computer Vision & Image Processing to detect and estimate 3D poses of objects in the scene. The technologies target challenging industrial use cases, bin picking with support for a large breadth of object types (rigid, flexible) and materials (matte, shiny).
- Object Detection & Pose Estimation (ODPE): Deep Learning approaches to compute the 3D poses of objects in the scene.
React: The Motor Cortex of Robotic System
The team pursues to improve on Robot Control & Grasping capabilities and functionalities to minimize setup time and improve user experience.
- Motion Optimization: Machine Learning algorithms to optimize robot control parameters to reduce setup time & minimize the need of user expertise while increasing throughput.
- Robot Grasping: Deep Learning algorithms to predict grasp points via virtual simulations.
