★ Safe and High-Throughput Intersection Protocols for Autonomous Vehicles [CMU]

Self-driving vehicle technologies are expected to play a significant role in the future of transportation. One of the main challenges on public roads is the safe cooperation and collaboration among multiple vehicles using inter-vehicle communications. In particular, road intersections are serious bottlenecks of urban transportation, as more than 44% of all reported crashes occur within intersection areas. To improve road safety and traffic throughput, we present a Configurable Synchronous Intersection Protocols (CSIP) and a Distributed Synchronous Intersection Protocols (DSIP). CSIP improves road safety by accounting for GPS inaccuracy, control failure, and time-synchronization failure, which are very common in the real-world systems. DSIP provides a decentralized mechanisms and mode-based decision-making policy, where each connected vehicle negotiates, controls, and enter the intersection. By using the dynamic decision-making framework, autonomous vehicles' behaviors in front of human-driven vehicles become very straightforward and comfortable, and never mislead the surrounding human drivers.

    • IEEE RTCSA 2019, V2V-based Synchronous Intersection Protocols for Mixed Traffic of Human-Driven and Self-Driving Vehicles. Best Paper Nominee

    • ACM Transactions on CPS 2019, CSIP: A Synchronous Protocol for Automated Vehicles at Road Intersections.

    • IEEE RTCSA 2017, A Configurable Synchronous Intersection Protocol for Self-Driving Vehicles.

★ Decentralized Cooperative Protocols for Dynamic Intersections [CMU]

Road intersections controlled by traffic lights and stop signs can be considered to be "stationary intersections" since they are captured in map databases, are known a priori, do not move, and last for a very long time. On the other hand, "dynamic intersections" might appear almost anytime and anywhere on public roads and might lead to potential dynamic conflicts that must be safely resolved. A dynamic intersection (a) represents a shared region on the road where traffic conflicts can arise, (b) may be temporary, (c) may arise dynamically due to the occasional presence of vehicles, and (d) is not included in a map database (unlike a multi-way road intersection or a roundabout). We present a decentralized cooperative protocol for dynamic intersections that can be used by self-driving vehicles for safely coordinating with other vehicles. Under our protocol, self-driving vehicles can also create a vehicular communication-based traffic manager named Cyber Traffic Light when the area is congested. A cyber traffic light functions as a self-optimizing traffic light by estimating the traffic volumes and by wirelessly coordinating among multiple self-driving vehicles.

    • ACM/IEEE ICCPS 2018, Dynamic Intersections and Self-Driving Vehicles.

    • ACM/IEEE ICCPS 2017, A Merging Protocol for Self-Driving Vehicles.

★ Co-Simulation Environment for Eco-Autonomous Driving [CMU]

Energy efficiency is one of the most significant factors for vehicles. However, testing, verifying and validating energy-efficient autonomous driving systems are difficult due to safety, cost and repeatability. To address this issue, we present a co-simulation platform to develop and test novel vehicle eco-autonomous driving technologies named InfoRich, which incorporates the information from on-board sensors, V2X communications, pre-loaded vehicle-related table and map database. The co-simulation platform includes eco-autonomous driving software, vehicle dynamics and powertrain (VD&PT) model, and a traffic environment simulator. Also, we utilize synthetic drive cycles derived from real-world driving data to test the strategies under realistic driving scenarios. To build road networks from the real-world driving data, we develop an Automated Parser and Calculator for Map/Scenario named AutoPASCAL. Overall, the simulation platform provides a realistic vehicle model, powertrain model, sensor model, traffic model, and road-network model to enable the evaluation of the energy efficiency of eco-autonomous driving.

    • IEEE IV 2020, Co-simulation Platform for Developing InfoRich Energy-Efficient Connected and Automated Vehicles.

    • Proceedings of the IEEE 2018, Tools and Methodologies for Autonomous Driving Systems.

★ Cooperative and Energy-Efficient Autonomous Driving Strategies [CMU, GM]

Eco-autonomous driving strategies include four eco-driving applications: Eco-Approach, Eco-Departure, Eco-Cruise and Eco-Lane Selection. These eco-autonomous driving strategies can be tested, designed and verified by using the co-simulation platform with the realistic vehicle model and realistic road networks. Eco-Approach is defined as the fuel-efficient vehicle/powertrain operation that applies coasting to bring a vehicle to stop in fuel-friendly fashion. Also, the Eco-Departure maneuver should conduct a smooth and less-aggressive acceleration profile to reach the target cruise speed. The target speed and the distance to reach the speed typically depend on the traffic preview information that are available from on-board sensors and vehicular communications. We developed these eco-driving strategies and are working on the implementation in the CMU self-driving software to achieve energy-efficient automation.

    • IEEE IV 2020, Co-simulation Platform for Developing InfoRich Energy-Efficient Connected and Automated Vehicles.

    • ASME DSCC 2020, Speed Trajectory Generation for Energy-Efficient Connected and Automated Vehicles. (Co-authored paper)

    • ARPA-E NEXTCAR Project: InfoRich VD&PT Control.

★ Deep Reinforcement for Cooperative Perception and Connected Vehicles [Toyota R&D InfoTech]

Cooperative perception enhance road safety by having connected vehicles exchange their raw or processed sensor data with the neighboring vehicles over Vehicle-to-Vehicle (V2V) communications. In fact, sensor-data capturing for the blind spot is helpful to avoid vehicle collisions and deadlocks. However, it heavily relies on V2V communications and generates negligible amount of data traffic. The excessive network load would increase the risk that important data packets are delayed or even lost. To keep the communication reliability by saving network resources, we present 2 frameworks: (i) Deep reinforcement learning for data selection and (ii) Data fusion model for avoiding information flooding. By using the deep reinforcement learning framework, each connected vehicle can intelligently identify pieces of perception data worth to transmit. To evaluate our frameworks, we designed and developed a Cooperative & Intelligent Vehicle Simulation (CIVS) Platform, where we integrate multiple open source software to constitute a unified framework, including a traffic model, vehicle model, communication model, and object classification model.

    • IEEE IV 2020, Cooperative Perception with Deep Reinforcement Learning for Connected Vehicles.

★ Participatory Sensing & Privacy for City Monitoring [UTokyo]

Participatory sensing is a crowd-sourcing approach to collect data by using participants' mobile devices. To show the feasibility, we designed and developed 2 components: (i) Participatory sensing framework and application and (ii) Perturbation-based privacy mechanism. First, we developed an application on Android OS that keeps sensing and storing sensor data, such as location, speed, sound level and air pressure. By collecting these data from city area, we can analyze and understand the city in a more fine-grained way. Secondly, since participatory sensing uses participants' devices, the applications have to preserve the privacy. To address the issue, we extended Negative Surveys and applied it to participatory mobile sensing. We developed and deployed the mobile application and visualized noise pollution maps in the central part of Tokyo.

  • IEEE ICC 2016, Democratic Privacy: A Protocol-hidden Perturbation Scheme for Pervasive Computing.

  • IEEE ICC 2014, Privacy-Preserving Community Sensing for Medical Research with Duplicated Perturbation.

  • IEEE CloudNet 2012, Limited Negative Surveys: Privacy-Preserving Participatory Sensing.

★ Bus GPS Data Mining for City Monitoring and Analysis [Microsoft Research]

Large-scale events attracting participants in urban area might have a strong negative impact on productivity, mobility, comfort and safety. To alleviate serious traffic jams and congestion, predicting the event occurrence beforehand is essential. When we know an event occurrence in advance, some of those who are not interested in the event might change their plans and/or might take a detour to avoid to get involved in a heavy congestion. In this context, we present an early event detection technique using GPS trajectories collected from periodic-cars, which are vehicles periodically traveling on a pre-scheduled route with a pre-determined departure time, such as a transit bus, shuttle, garbage truck or municipal patrol car. These trajectories provide the real-time and continuous traffic in city area without incurring any privacy invasion. To detect the events, we use graph-based anomaly detection techniques and Time-dependent Congestion Network. We evaluate the methodologies with over 7,000-bus data collected in 2015 at Beijing, China.

  • ACM SIGSPATIAL 2017, An Early Event Detection Technique with Bus GPS Data.

  • IEEE Transactions on Big Data 2019, BusBeat: Early Event Detection with Real-Time Bus GPS Trajectories.