Hardware Podcasts

79. Teaching Robots To Find Buried Objects

The Next Byte Podcast · 19 Jul 2022 · 30m

AI in practice: one real use of AI described

What this episode covers

Hosts Daniel and Forbode discuss MIT research on FuseBot, a system using RFID and visual mapping to help robots find buried objects, and University of Michigan research on using fake data to accelerate robot training. They detail how FuseBot achieves 95% retrieval success and 4x efficiency when items are tagged, while the University of Michigan team demonstrates 40x faster training and 45% higher success rates for cobots by injecting human-like generalizations into training data. The hosts explore applications in warehouse logistics, search and rescue, and construction, highlighting the shift from brute-force physical trial-and-error to intelligent, data-augmented learning.

AI in practice

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

Engineering & simulationFuseBot (RF visual mapping and extraction system)95% success rate vs 84% for X-ray; 40% more efficient retrieval

The system used RF and visual data to create a 3D occupancy distribution map, optimizing the robot's path to retrieve buried items by minimizing steps and recalculating probabilities as items were moved.

“Fusebot was able to retrieve 95% of partially or fully covered target items in comparison to x-rays 84% success rate. Now here's where it gets even more interesting. In the instances where they were both able to be successful in retrieving the target item, FuseBot was 40% more efficient, and that efficiency increases with the number of RF items that are tagged in the pile.”

Speaker 1 · Hear it at 10:18

Most relevant to

early-stage companies

Topics

Mentioned

Companies
MIT, University of Michigan
Products
FuseBot, X-ray, Boston Dynamics Spot
Chips
RFID

Related episodes from other shows

Other shows on the same ground, closest first.

  • Embedded.fm · 55m

    377: Robot at the Park

    Erin Kennedy, founder of Robot Missions and known as RobotGrrl, discusses her work building robots to clean plastic from shorelines and her involvement in the Hackaday Dream Team project to reduce ghost fishing gear. She details the technical challenges of her 'Bowie' robot, including navigating sand, using TensorFlow for debris detection, and the limitations of GPS versus AR markers for localization. Kennedy also shares insights on her unconventional learning path, her part-time role at an innovation lab, and her recent 'fog voxel' hobby project using piezo discs to create volumetric pixels.

    Robotics & autonomyAI & machine learning
    10 chapters

  • The Robot Report Podcast · 1h 17m

    Ken Goldberg on robotic grasping; Carbon Robotics’ new weeding robot

    Ken Goldberg and Jim Liefer of Ambi Robotics discuss their Sim2Real AI approach to robotic grasping, explaining how they trained neural networks on simulation data to handle diverse e-commerce objects without human intervention. Paul Mikesell, founder of Carbon Robotics, details the development of their third-generation autonomous weeding robot, which uses high-powered lasers and computer vision to identify and destroy weeds in row crops. The episode also covers WTWH Media's acquisition of Mobile Robot Guide and updates on the Mars 2020 Ingenuity helicopter.

    Robotics & autonomyAI & machine learning
    7 chapters

  • Robot Talk · 29m

    Episode 130: Robots learning from humans - Chad Jenkins

    Chad Jenkins, a Professor of Robotics at the University of Michigan, discusses his lab's work on teaching robots through demonstration rather than coding, focusing on dexterous mobile manipulation and perception. He explains the challenge of 'affordances'—understanding how objects can be used—and the trade-off between the speed of deep learning models and the robustness of search-based algorithms. Jenkins also highlights his 'distributed teaching collaboratives' initiative, which uses open-source course materials to scale robotics education, supported by funding from the Sloan Foundation, Amazon, and Toyota Research Institute.

    Robotics & autonomyAI & machine learning
    7 chapters

  • The Robot Report Podcast · 1h 04m

    Baseball & robotics startups during COVID-19; ROS on Windows 10

    AJ Mayer, CEO of the Boston-based Pickle Robot Company, explains how his startup uses reinforcement learning and deep neural networks to automate loading and unloading trucks with their 'Dill' robot, targeting brownfield logistics facilities rather than fully automated warehouses. He details the 'supervisor ratio' business model, where human operators handle the 2% of edge cases the robot cannot manage, and discusses the challenges of selling to enterprise customers during the pandemic. The episode also features Dave Niewinski from Clearpath Robotics, who discusses the strategic benefits of supporting ROS on Windows 10 for their Unmanned Ground Vehicles (UGVs), highlighting easier IT integration, cloud connectivity via Azure, and lower barriers to entry for corporate clients.

    Robotics & autonomyAI & machine learning
    8 chapters

  • The Robot Report Podcast · 1h 05m

    From Teleoperation to Autonomy: Inside Boston Dynamics' Atlas Training

    Scott Kuindersma, VP of Robotics Research at Boston Dynamics, explains how the company trains the Atlas humanoid robot using Large Behavior Models (LBMs) and teleoperation data. The team collected 20 hours of teleoperation data to enable Atlas to perform complex bi-manual manipulation tasks, such as picking and placing Spot parts, while adapting to real-time disturbances. Kuindersma details the four-stage pipeline of data collection, annotation, model training, and evaluation, highlighting the critical role of simulation data and human demonstrations in generalizing these skills. The episode concludes with plans to test Atlas in Hyundai facilities and further explore AI-centric dynamic behaviors.

    Embedded firmwareRobotics & autonomy
    10 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.