Infrastructure-less ITS with next-generation devices

MIT Investigator(s): Li-Shiuan PEH (LEAD)

Team Members:
  • Pilsoon CHOIResearch ScientistMIT-CSAIL/EECS
  • Jason GAOPhD StudentMIT-CSAIL/EECS
  • Woo-Cheol KWONPhD StudentMIT-CSAIL/EECS
  •   Collaborators
  • Tulika MITRAProfessorNUS
  • Mun Choon CHANAssoc. Prof.NUS
  • Weng-Fai WONGAssoc. Prof.NUS
  • Kartik SANKARANPostdocNUS
  • Most ITS systems today require the deployment of costly physical roadside infrastructure such as gantries, traffic signals, signs, and sensors embedded within the fixed transportation infrastructure. As a result, deployment and maintenance of ITS systems remains highly costly, and tends to be limited to selected regions rather than island-wide. Next-generation devices will comprise sufficient computing, networking, sensing hardware to enable the realization of truly infrastructure-less ITS, realized entirely with on-board or mobile devices/wearables.

    Future ITS Applications with Next-Generation Wearables
    • The objective is to design and prototype next-generation ITS leveraging novel platforms of wearables with low-power sensors: (1) Develop and deploy transport applications on wearables (e.g. smart watches) that possess just ultra-low-power sensors (e.g. step counter, baro, gyro, accelerometer). Research challenges arise in detecting transport modes and activities without conventional, power-hungry location sensors such as GPS. (2) Architect next-generation many-core wearable processor chips that can enable the above transport applications at truly real-time performance. Research challenges arise as today’s wearable processors cannot support true real-time, in-situ processing; some analytics has to be deferred for off-line processing.
      Future ITS Applications with Next-Generation Phone Comms and Sensors
    • The objective is to design and prototype next-generation ITS leveraging next-generation phone-based LIDAR and V2X radio hardware. We recently developed a smartphone-based LIDAR sensor that attaches a low-cost laser to phones, and uses the smartphone’s camera and processors for constructing a LIDAR terrain map [ICRA 2016]. We have done preliminary testing of our smartphone-based LIDAR as a collision avoidance sensor on the SMART autonomous vehicle. We will further extend that to full system trials with vehicle control in concert with the phone-based LIDAR sensor, demonstrating that phones can be ultra-low-cost, pervasive sensors enabling rapid deployment of driverless vehicles.
    • In FM1, we have proposed and developed system prototypes of V2X-on-phones, and demonstrated several ITS applications that will be enabled by such phone-to-phone comms, such as SignalGuru, RoadRunner, etc. Such V2X-on-phones enable the connectivity of not just vehicles, but pedestrians and passengers. Now, we propose to push these V2X-on-phones radios further towards production, miniaturizing them to the form factor of dongles that can be connected to existing smartphones. This will enable, for the first time, truly mobile experiments, as the previous system prototypes are huge and can only permit lab-based experiments. We have been working with industry to fabricate these dongles. Our plan is to deploy them to MIT students on campus, as well as at Singapore. In Singapore, these V2X-on-phone radio dongles can be deployed in concert with the wearables and phones in the earlier project, as well as with the phone-based LIDAR sensors on AVs.
      Multiple Sensor Data Feeds
    • The objective is to identifying transport feature signatures from multiple data streams. While each sensor (either infrastructure sensors in a Intelligent Transportation System or infrastructure-less sensors in a mobile handheld) provides a stream of data where interesting feature signatures can be recognized, the power of such a system is in the aggregation of multiple feeds. One of the main questions in multiple feeds is the movement of interesting feature signatures (people, cars etc.) across multiple feeds to trace a path. Furthermore, some feature signatures become important due to the movement (car moving too fast between points). However, finding the signature correlations across multiple feeds is very hard as each signature can be small and fleeting and it is hard to know what sensor feed the signature will appear next and when that will occur.