tutorial

RoSÉ Tutorial at MICRO 2023:

Designing, Deploying, and Evaluating Full-Stack Robotics Systems With RoSÉ

Overview

We are running a half-day tutorial on RoSÉ, an open-source hardware-software co-simulation infrastructure that enables the evaluation of robotics system-on-chips (SoCs).

Using RoSÉ, users will be able to deploy end-to-end simulations of a robotics SoC running in a simulated environment. By co-simulating a cycle-exact RTL simulation with a robotics simulator, users can evaluate the performance of their SoC, capturing performance metrics of both CPU cores and hardware accelerators, and gain system level insights about architectural choices. Users will also be able to evaluate how their SoC can perform in differing maps, sensor configurations, and tasks by using Unreal Engine and AirSim to construct a robot’s simulated environment. Finally, users will be able to explore different software choices for their robotics application, including both DNN-based and classical robotics workloads.

By completing this tutorial, users will be able to use RoSÉ to evaluate the end-to-end performance of their robotics SoC through customized experiments.

A tentative schedule is available below.

Logistics and Registration

Venue: MICRO 2023 (will be held in-person)

Date: Sunday, October 29th, 2023

Time: 8 AM - 12 PM EDT

Registration

  • Please register to attend MICRO.
  • Additionally, please fill out this Google form if you would like to attend our tutorial.

Schedule (Tentative)

Introduction to RoSÉ

Participants will be introduced to the RoSÉ framework and its capabilities for hardware-software co-simulation in robotics.

This section will also cover:

  • A high-level overview of RoSÉ’s architecture and interfaces.
  • The scope of cyberphysical systems that can be modeled with RoSÉ.
  • Hardware and software requirements needed to use RoSÉ.

Building SoC Designs

Participants will:

  • Learn how to select an appropriate SoC configuration for their robotics application.
  • Be guided on selecting between SoC configurations provided by RoSÉ.
  • Get an overview of generating new configurations using Chipyard.
  • Learn to deploy a bitstream to an FPGA using RoSÉ, generated by FireSim.
  • Learn how to configure the sensor interfaces supported by RoSÉ.

Building Robotics Environments

Participants will:

  • Learn to create environments using AirSim for robotics experiments.
  • Configure sensors, actuators, and robot models.
  • Build a simulation executable.

Building a Software Pipeline

Participants will:

  • Learn to build robot software pipelines using ONNX Runtime and ROS (Robot Operating System).
  • Exlore learning-based workloads in RoSÉ
    • Learn Methodologies for mapping the DNN model to hardware accelerators in RoSÉ, such as Gemmini.
    • Evaluate Hardware/software co-design optimizations such as quantization.
    • Integrate hardware schedulers to support the QoS targets of robotics applications.
  • Evaluate end-to-end applications using ROS

Simulation Deployment and Experimentation

Participants will:

  • Learn to configure and launch simulations using RoSÉ, evaluating hardware and software design-spces.
  • Extract architectural performance metrics and robot sensor logs.
  • Gain insights into analyzing and evaluating their robotics application’s performance.
  • Discuss the trade-offs of tuning RoSÉ’s synchronization granularity.