Home

JUMAX Access and Webots Installation

The following pages explain how to access to the JUMAX machine, how to compile and start Webots on it and also expose a benchmark of the 2 Webots display options.

• How to Access Jumax
• How to Compile Webots on Jumax
• How to Start Webots

MaxIDE and MaxCompiler

The following pages explain the basics of accelerated applications, how to start the Maxeler IDE to code DFE applications (MaxIDE), and how to compile them using MaxCompiler.

• Basics of DFE Applications
• Start MaxIDE
• Run DFE Applications with MaxIDE
• Compile DFE Applications for Webots
• Compilation Debug and Timing Improvement

Accelerated Multilayers Perceptrons

In order to know the order of magnitude of the gain that can be obtained on the Jumax FPGAs, the following workflow is applied:

• Implementation of multilayers perceptrons on CPU :
  • Creation of a MLP Deep Learning Framework in C++
  • MLP Forward Propagation on CPU: Tests and Results
• Adapting the multilayers perceptron (MLP) inference for FPGA:
  • MLP Forward Propagation on DFE: Structure
  • MLP Forward Propagation on DFE: Results and Optimization

Autonomous car simulation using Convolutional Neural Networks

The following steps show the implementation of a state-of-the-art robotic application in simulation which uses a Convolutional Neural Network (CNN) and highlights the difference in performance between CPU and FPGA in this context.

• Implementation of a state-of-the-art robotic application in simulation
  • Literature review
  • Trajectory planning and dataset collection
  • Neural network architecture and training
  • Inference in simulation
• Convolutional Neural Network inference on CPU
• Convolutional Neural Network inference on FPGA
• Deliverable 1: CNN performance comparison CPU/FPGA
• Deliverable 2: CNN performance comparison CPU/FPGA
• How to run the most optimized car simulation