Hardware Podcasts

195. Wear This Exoskeleton & Save Your Knees!

The Next Byte Podcast · 8 Oct 2024 · 21m

AI in practice: one real use of AI described

What this episode covers

Hosts Daniel and Forbode discuss a University of Michigan research project developing a knee-focused exoskeleton to support blue-collar workers. The device retrofits commercial knee braces with drone motors and uses a physics-based machine learning model to assist the quadriceps during repetitive lifting. The hosts highlight test data showing a 44% performance retention in fatigued users and discuss the potential for reducing workplace injuries and increasing productivity through affordable, scalable hardware.

AI in practice

Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.

Engineering & simulationphysics-based machine learning model124 updates per second

The team used a physics-based machine learning model to predict leg kinematics and adjust motor support in real-time, reducing latency from one second to 124 updates per second.

“these folks decided to use a physics-based machine learning model, which means, you know, it has algorithms running physics equations, basically, that can tell it, you know, the kinematics of, if I'm moving my leg this way, it probably means this is the trajectory I'm going. So, you know, adjust with me. Don't fight against me when I'm doing that. And they said it's updating about 124 times per second.”

Speaker 1 · Hear it at 11:06

Most relevant to

early-stage companies

Topics

Mentioned

Companies
Mouser Electronics, University of Michigan, Wevolver

Related episodes from other shows

Other shows on the same ground, closest first.

  • Over the Air · 29m

    From Wearables to Insurance: How Kinetic is Revolutionizing Workplace Safety

    Heytham, co-founder and chief strategy officer of Kinetic, explains how the company evolved from designing wearable IoT devices to becoming a workers' compensation insurance provider. He details the hardware development of a belt-mounted sensor using Intel Edison chips to monitor body mechanics and prevent musculoskeletal injuries in frontline workers. Brock describes the strategic pivot to insurance to serve smaller companies, leveraging partnerships with insurers like Nationwide to offer risk-reduction technology alongside coverage. The episode highlights the use of computer vision and mobile apps to identify facility risks, focusing on industries like last-mile delivery and warehousing.

    Industrial & consumer product designMicrocontrollers & dev platforms
    9 chapters

  • The Robot Report Podcast · 1h 10m

    Sarcos’ Ben Wolff, Uber ATG low on fuel, ROS on Spot

    Ben Wolff, chairman and CEO of Sarcos Robotics, discusses the development and commercialization of the Guardian XO full-body powered exoskeleton, highlighting its ability to reduce power consumption from 6,800 watts to 500 watts and its 'Robot as a Service' business model. The episode also covers Uber ATG's financial struggles, noting its $50 million annual burn rate and recent funding pressures, and features Dave Niewinski of Clearpath Robotics explaining their ROS package for the Boston Dynamics Spot quadruped, which enables developers to integrate the robot into existing ROS ecosystems for tasks like disaster relief and inspection.

    Mechanical engineeringStarting a product company
    10 chapters

  • Embedded.fm · 1h 12m

    187: Self-Driving Arm (Repeat)

    Patrick Pilarski, a Canada Research Chair in Machine Intelligence for Rehabilitation at the University of Alberta, discusses the development of smart prosthetic limbs that use machine learning to restore function for amputees. He explains how reinforcement learning allows prosthetics to learn from trial-and-error or by observing a user's healthy limb, addressing the challenge of mapping limited biological signals to complex robotic movements. Pilarski highlights the shift from traditional myoelectric controls to systems that predict user intent and context, such as using cameras to identify objects like soup or doorknobs to automate grip patterns.

    Embedded firmwareAI & machine learning
    7 chapters

  • Robot Talk · 51m

    Episode 90: Robotically Augmented People - Robot Talk Live

    Milia Hasbani of Imperial College London, Benjamin Metcalfe of the University of Bath, and Dani Clode of Cambridge University discuss robotic prosthetics and human augmentation. Hasbani explains her active prosthetic arm research using EMG and computer vision to select grasp types, while Metcalfe details a hybrid exoskeleton combining robotic support with functional electrical stimulation for rehabilitation. Clode presents her 'Third Thumb' and the Alternative Limb Project, exploring neuroplasticity, embodiment, and the ethical implications of augmenting able bodies.

    Medical & health devicesSensors & measurement in products
    10 chapters

  • Embedded.fm · 1h 12m

    187: Self-Driving Arm

    Patrick Pilarski, a Canada Research Chair at the University of Alberta, discusses his work on smart prosthetic limbs that use machine intelligence to bridge the gap between limited biological signals and complex robotic actuators. He explains how reinforcement learning allows prostheses to learn from trial and error in real-time, using techniques like 'prosthetic falconry' where a robotic limb learns by observing a user's healthy biological arm. Pilarski highlights the shift from simple myoelectric controls to systems that predict user intent and context, such as using cameras to identify objects like soup or doorknobs to pre-shape grips. He also outlines accessible tools for engineers, including Weka, OpenIA Gym, and Robotis Dynamixel servos, to get started with reinforcement learning robotics.

    Embedded firmwareRobotics & autonomy
    8 chapters

The summary, chapters and topics are generated from the episode’s audio by this directory, not written by The Next Byte Podcast. The episode itself belongs to its publisher.