Intern - France

Internship – Pedestrian Dead Reckoning using IMUs Explore Recent State of the Art Deep Neural Networks Solutions 
Develop a Demo
 

We are looking for A motivated student to join our Algo Motion team at TDK/Invensense the world’s leading provider of MEMS sensor platforms (accelerometers, gyroscopes, magnetometers, and microphones). You will join our small and friendly SW team, located in Grenoble city, close to the train Station.

 Overview

At TDK/InvenSense, we’re creating Algorithms for consumer devices (SmartMotion® | TDK InvenSense). 
We so add meaningful functionalities from counting steps to navigation capabilities to devices such as Smartphones, Smartwatches, Rings, Earbuds and Smart Glasses.

TDK/InvenSense is delivering Inertial Measurement Units (IMUs) combining an accelerometer and a gyroscope and on board SW with Algorithms, and Magnetometers, to Device Manufacturers. The IMU comprises a tiny Microcontroller that can embed SW carrying our Algorithms.

For now quite a long time, one of the missions of the IMU (+ option Magnetometer) has been to help its user to navigate in indoor/outdoor environments. The Smartphone Map Application is commonly used every day by pedestrians esp. in cities to find their destination and the IMU data participates to navigation.

For Outdoors, Pedestrian Navigation is mostly very successfully relying on Satellite positioning. In Urban Canyons, GPS can become unreliable. For Indoors, Satellites are not operational. Other sources allowing Positioning like Wi-Fi, BLE, Cellular or Vision Based Location all suffer from limitations like dependency to Databases, equipment, infrastructure, and can be costly and power hungry. IMU are cheap, low power, present in all Smartphones, e-Watches, most Earbuds and all Smart Glasses. The more the IMU can help Navigation, producing a Pedestrian Odometry (Distance and Heading) at low cost, the more the other sources can be duty cycled or power off, and the less the navigation will depend on external infrastructures.

Pedestrian Odometry using IMUs (also called Pedestrian Dead Reckoning “PDR”) has been a hot topic for quite some years, and we have been using the most modern Algorithm techniques up to the 2020s. Deep Neural Net appear to offer new capabilities and we would like to explore these to compete with our traditional PDR solution.

Internship Description

At InvenSense Grenoble we have been using Machine Learning techniques for some time to create new Algorithms in the field of Detection/Classification, such as for Detecting/Classifying a Gesture of the User (eg Yes / No) or Classifying the State of a System (such as Wearing Smart Glasses or Not). We are testing, in R&D, the approach from real sensor data acquisition to some annotated outputs for several use cases, like Accelerometer based “Bring to See” to wake the watch when the user brings it to his attention, "Wrist Gestures" to navigate in the menus of an e-watch, "Head Gestures" to control smart glasses for incoming calls, "Voice Detection" for smart glasses through Accelerometer based bone conduction …

We want to test Deep Neural Net methods, such as found in recent literature to beat our traditional PDR.

1. Evaluation of An Indoor Localization Engine
   Authors: Christophe Villien, Anne Frassati, and Bruno Flament
   Publication / Venue: 2019 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Pisa, Italy (IEEE)
   Year: 2019
2. RoNIN: Robust Neural Inertial Navigation in the Wild: Benchmark, Evaluations, and New Methods
   Authors: Hang Yan, Sachini Herath, and Yasutaka Furukawa
   Publication / Venue: IEEE International Conference on Robotics and Automation (ICRA) (arXiv preprint 2019)
   Year: 2019 / 2020
3. IMUNet: Efficient Regression Architecture for Inertial IMU Navigation and Positioning
   Authors: Behnam Zeinali, Hadi Zanddizari, and Morris J. Chang
   Publication / Venue: IEEE Transactions on Instrumentation and Measurement (Vol. 73, Art no. 2516213)
   Year: 2024
4. FTIN: Frequency-Time Integration Network for Inertial Odometry
   Authors: Shanshan Zhang, Qi Zhang, Siyue Wang, Liqin Wu, Tianshui Wen, Ziheng Zhou, Ao Peng, Xuemin Hong, Lingxiang Zheng, and Yu Yang
   Publication / Venue: Research Publication / Conference Proceedings
   Year: 2025
5. EqNIO: Subequivariant Neural Inertial Odometry
   Authors: Royina Karegoudra Jayanth, Yinshuang Xu, Ziyun Wang, Evangelos Chatzipantazis, Kostas Daniilidis, and Daniel Gehrig
   Publication / Venue: International Conference on Learning Representations (ICLR)
   Year: 2025

We want to explore the capabilities of such DNN approaches on our Data and or using the available Public Databases.

We are a Team of 6 Algorithm People, experts at using TensorFlow (PyTorch welcome too), experts at knowing the best we can obtain from our IMUs with traditional Algorithm Design techniques, and we are producing some Algorithms that are found in many Smartphones, Wearables, Smart Glasses, …

The Intern would join a Team that will be happy to help. He / She would focus on:

• Prepare / Augment databases to feed Training and Testing the DNN based PDR.
• Read Papers, and with the Help of our Experts, select a reasonable, rationale approach to test the Deep Neural Net Concept for PDR Application,
• Design/Train Dense Deep Neural Network. We are using TensorFlow (PyTorch also welcome), so scripts and help will be available to ease the process, 
• Test the So created Algorithm using our existing Metrics (that we have for our existing Traditionnal PDR)
• We also want to improve / design a Real Time Demo running on a PC, that we want to promote our Solutions.

Another purpose of the internship is to build an efficient and ergonomic App for our Pedestrian Dead Reckoning algorithm. Using graphical tools helps to improve development and achieve better performance. Firmware team has built Sensor stage app to provide a common tool to help other teams as Algo Motion team to quickly test and tune their own algorithm. It is an internal tool developed in Python using PySide6 framework. We will so also devote some time to build this nice dependable App. 
 

Qualifications

• 3rd year of Engineering school

• Programming and optimization in Python, use Git. Some C Code.
• Knowledge and interest in machine learning methods and esp. Deep Neural Nets
• Taste for testing new Algorithms, Manipulate Data, Plot Data,
• Some knowledge or interest in sensors and/or 3D geometry.
• Desire for running experiments with sensor prototypes.
• Pragmatic mindset, autonomy, and creativity.
• Pugnacity and creativity
• Duration: 6 months
   

InvenSense, a TDK group company, is a world-leading provider of MEMS and magnetic sensor solutions found in mobile, wearable, smart home, AR/VR, and other consumer and IoT applications. We are innovators with a 20+ year history of patented sensor technologies. As experts in our field, we lead industry breakthroughs in consumer, industrial and automotive technology and reliably deliver high-quality solutions to our partners across global markets.Learn more about us at invensense.tdk.com