2012/fpaa12

Developing autonomous flying robotics using Neuromorphically Inspired Analog-Digital Processing

Members: Asha Gopinathan, Christian Brandli, Christoph Maier, Daniel B. Fasnacht, Magdalena Kogutowska, Mostafa Rahimi Azghadi, Mathis Richter, Sadique Sheik, Sudarshan Ramenahalli, Tara Julia Hamilton, Tobi Delbruck, Terry Stewart, Theodore Yu

Leaders: Scott Koziol, 'hasler'

The focus of this project is to integrate a reconfigurable analog processor capable of implementing neuromorphic based AVLSI algorithms with a robot for Navigation, Guidance and Control (NG&C). Those involved in this project can be involved at multiple levels of the AVLSI Hardware/Software/Robot? co-design problem. As shown in the table below, Guidance is the process of determining a path for the robot to reach the goal, Navigation is determining the robot’s state such as position, velocity and attitude, and Control is tracking guidance commands while maintaining stability.

Function Standard Methods
Guidance Path Planning A* (pronounced A-star)

 Navigation Determines state Kalman Filter

Control Tracking / stability PID controller

The group direction will focus on a mobile air platform will be a combination of three items, the robot hardware, avionics, and sensors. The robotic hardware is a quad-rotor helicopter platform; we see this platform as a stable first approach before moving to small or micro aerial vehicles (MAV). The avionics envisioned includes processor hardware that includes a board stack with an FPAA, Atmel microprocessor and a suite of sensors including MEMs based gyros and accelerometers, and a VGA camera. These capabilities add to, or could replace, capabilities on the copter. Multiple FPAAs could be integrated together. Additional sensors may be able to be integrated, and the organizers would encourage such discussions.

A fundamental perspective will be to utilize the computational power efficiency (x 1000 over digital algorithms) and area efficiency of analog hardware to enable low-power autonomous flying platforms. We expect the use of neuromorphic approaches will further improve the resulting efficiency and performance of these algorithms, and in this group, we will attempt to make the first steps in this direction.

Because of the unique approach enabled by the FPAA device, most of the effort will be in embedded software design that almost certainly can include analog circuit design.

The organizers welcome involvement in some or all of the resulting areas, and would encourage suggestions in this approach.

Hands-on Overview of FPAA Chips and Tools:

A significant part of this project will involve Large-Scale Field Programmable Analog Arrays (FPAA), technology invented and used at GT and used in multiple labs at this point. FPAAs, like FPGAs enables a configurable approach to analog and mixed signal approaches typical of digital systems (i.e. FPGAs, uP). Many important aspects of neuromorphic design can be implemented in physical approaches; therefore having such techniques makes these device, circuit, and system approaches accessible to a wider audience. This tutorial will introduce in the theory, chips, boards, and tools over hands-on sessions approach. These approaches provide a useful framework for discussing where to use neuromorphic type design approaches in a range of applications. We can also discuss related topics to these approaches, including programmable (floating-gate) circuits, that enable memory, programmable devices, and adaptive devices in a dense, low-power way into our neuromorphic systems. We welcome discussions that integrate concepts / projects from other workgroups / tutorials.

Building Block Projects

Analog--Digital Subsampling Image Convolution using FPAAs + VGA Imager

Integration using the VGA imager already integrated on our board stack doing a low-speed subsampling image convolution using the FPAA IC. The project will be a hands on project to learn the FPAA circuitry and capability by integrating with a commercial sensor device.

Other projects

Other approaches could include using the FP A A for processing the outputs of the MEMS accelerometers and gyroscopes as part of an Inertial Navigation System (INS), obstacle avoidance, and hover height control using the FPAA sensors

Larger Goal Projects

A proposed large project goal is to try to get the airborne device as an autonomous system that can hover and follow someone based on visual cues and keeping the copter stable.

Papers / Materials to review

Attachments