Predictive Data-Driven Vehicle Dynamics and Powertrain Control: from ECU to the Cloud

Our goal

We are developing a predictive and data-driven control technology for a connected and automated plug-in hybrid electric vehicle. Our goal is to reduce the total vehicle energy consumption by at least 20%. We use historical and real-time data-feeds to co-optimize powertrain and vehicle dynamics control. We coordinate real-time predictive control on the on-board control unit with cloud-based predictive control. We are working to demonstrate this technology in real-world driving.

Our goal is to harness the untapped potential of remote computations, forecasts, historical data, automation and coordination with other vehicles and infrastructure. We are developing a novel adaptive and predictive approach enabled by data, communication and cloud computations.

Our system

Vehicles

The ego-vehicle is the vehicle on which we evaluate the performance of our control approach. The fleet vehicles are other test vehicles that are used in our demos. They feature

  • hybrid and plug-in hybrid electric powertrains
  • radar and camera for environment perception
  • GPS for localization
  • DSRC (Dedicated Short Range Communication)
  • 4G connectivity

Infrastructure

We leverage smart signalized intersections instrumented by Sensys Networks. They feature

  • vehicle detection sensors
  • SPaT (Signal Phase and Timing) prediction
  • DSRC module to braodcast SPaT information
  • cloud connectivity for historical data logging

Cloud

We leverage the Berkeley hosted platform for data logging and real-time optimization.

Our demos

Our team includes three UC Berkeley laboratories, a major OEM and a silicon valley company focused on wireless traffic and transportation infrastructure data. We built a fleet of test vehicles and identified three demos in real-world driving scenarios.

Eco-approach/departure at signalized intersections

In this demo, the ego-vehicle drives autonomously through a corridor of signalized intersections.

The ego-vehicle constantly updates a map of the surrounding environment using the on-board sensors and the forecasts received via DSRC. On-board sensors detect other vehicles, cyclists, pedestrians, road signs and markings. Other fleet vehicles communicate their predicted trajectories. The traffic lights communicate the remaining duration of the current light phase. Using this dynamic map, the ego-vehicle avoids collisions and prevents energy-wasteful behaviors like idling at red lights, ill-timed engine switching, or excessive throttling.

On-board sensors and DSRC have limited range, but we can use cloud data and computations to make the ego-vehicle more long-sighted. In this demo, we use historical data of traffic light phases to plan in advance the future speed trajectory of the ego-vehicle.

Predictive eco-approach/departure at signalized intersections.

Predictive eco-CACC and speed harmonization.

Eco-CACC (Cooperative Adaptive Cruise Control) and speed harmonization

In this demo, the ego-vehicle and other fleet vehicles drive cooperatively on a highway.

The ego-vehicle and other fleet vehicles constantly update the map of their own formation (or platoon) and that of the surrounding environment. Vehicles in the platoon still use their on-board sensors to make sure they do not collide with each other and with the surrounding vehicles; they also exchange their future speed and acceleration via DSRC, thus they are able to drive with small gaps between each other. Driving in a platoon, they enjoy lower air resistance, reduce useless acceleration or deceleration, and increase the road throughput.

Also in this case, we can use cloud data and computations to work around the short sight of on-board sensors and DSRC. In this demo, we use forecasts of road grade, traffic speed, and traffic density to plan in advance the future trajectory of the platoon. For instance, the platoon can adapt its speed in advance if a traffic jam is on its way.

Learning-based eco-routing

In this demo, the ego-vehicle improves its routing choices by learning from its own past driving data.

With a perfect model of its own energy consumption and real-time data on the road network, the ego-vehicle could compute the energy-optimal route from an origin to a destination. In practice, model tuning can be inaccurate or deteriorate over time, and precise traffic data are hard and expensive to obtain.

In this scenario, the ego-vehicle constantly logs data like speed profiles and energy consumption, and pushes them to the cloud. Cloud-based learning constantly updates the vehicle model, and identifies patterns in the driving environment. Constant improvement of the vehicle model is useful for instance to mitigate the effects of vehicle aging. The driving environment has some features that are easy to identify or even retrieve from maps (like the road grade). For other features, instead, it is difficult or expensive to retrieve data from a third-party: for instance, it is hard to find the pattern of traffic speed at a particular intersection - unless driving data are collected over time.

Publications

Firoozi R, Guanetti J, Borrelli F (2018), "Vehicle Localization and Control on Roads with Prior Grade Map", Manuscript submitted to the 57th IEEE International Conference on Decision and Control , December 2018.

Guanetti J, Kim Y, Borrelli F (2018), "Control of connected and automated vehicles: state of the art and future challenges", Annual Reviews in Control. Article in press.

Guanetti J, Borrelli F (2018), "Optimal eco-routing of plug-in Hybrid Electric Vehicles", Manuscript in preparation.

Kim Y, Tay S, Guanetti J, Borrelli F (2018), "Hardware-In-the-Loop Setup for Connected Automated Vehicles Testing in Real Traffic", To appear at the 14th International Symposium on Advanced Vehicle Control, July 2018.

Kim Y, Tay S, Guanetti J, Borrelli F (2018), "Eco-driving with Learning Model Predictive Control", To appear at the 14th International Symposium on Advanced Vehicle Control, July 2018.

Kim Y, Zhang X, Guanetti J, Borrelli F (2018), "Robust Model Predictive Control with Adjustable Uncertainty Sets", Manuscript submitted to the 57th IEEE Conference on Decision and Control, Dec 2018.

Ibrahim S, Kalathil D, Sanchez R O, Varaya P (2018), "Estimating Phase Duration for SPaT Messages". Submitted to IEEE Transactions on Intelligent Transportation Systems.

Sun C, Shen X, Moura S (2018), "Robust Optimal Eco-driving Control with Uncertain Traffic Signal Timing". To appear at the 2018 American Control Conference.

Sun C, Guanetti J, Borrelli, Scott Moura (2018), "Data-Driven Chance Constrained Robust Optimal Eco-Driving Control of CAVs", Manuscript submitted to IEEE Transactions on Intelligent Transportation Systems.

Acknowledgement

The information, data, or work presented herein was funded in part by the Advanced Research Projects Agency-Energy (ARPA-E), U.S. Department of Energy, under Award Number DE-AR0000791. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.