Intelligent Cockpit using Vehicle and Driver Data

Intelligent cockpit concept 

Intelligent cockpit system concept

Traffic Estimation using Big Data Analytics

We use real data to model the uncertainty and congestion on road segments. The analysis of big data is essential for modeling the problem and verifying the validity and efficiency of the algorithms. The transportation data are abundant and need to be analyzed to capture descriptive information. Our algorithms address data analysis, inference, and estimates for traffic conditions using static and dynamic sensors. We enable real-time estimation of flow on roads with dynamic sensors calibrated by existing static sensors. Based on the real-time flow and speed information that we obtain from dynamic sensors, we predict the congestion levels, and we make it possible to identify the current operational state of the roads. These estimates are very valuable information for both individual drivers and city planners.

The traffic models for the algorithm are estimated using the real data from inductive loop detectors and a fleet of roving taxis, and they capture important characteristics of traffic conditions, including uncertainty in travel time and congestion created by agents’ road usage. We designed the predictive model to infer the general traffic flow from taxi data and loop detector data.

Multi-agent (vehicle, robot) Stochastic Path Planning


We have developed congestion-aware routing algorithms based on data-driven traffic modeling that considers uncertainties in travel time and the congestion effect according to the drivers’ path choices. The algorithms are designed in such a way that they can be implemented in a large network with a large number of agents. These highly efficient algorithms make possible much improved large-scale traffic planning.

Path planning for multiple agents has been studied extensively, but there are still many challenges in theory and practice. Current in-car navigation systems or Web-based route planning systems provide a driver a path from his origin to destination. However, oftentimes the path turns out to be congested. Avoiding congestion and finding a good path are not trivial, and the challenge is two-fold: (1) it is hard to predict the travel time for a road segment since many factors such as road conditions, congestion, and events affect the travel time, and (2) congestion is not fixed but rather is an aggregate effect of drivers’ choice of road segments to take for their trips. Thus, a driver's path choice should be optimized considering the other drivers’ path choices, and the decision about paths affects other drivers’ travel time at the same time.

Our first set of algorithms includes stochastic shortest-path planning to enable a driver to find a fast and reliable path. The algorithm copes with the uncertainty of road traffic conditions by stochastic modeling of travel delay on road networks. The algorithm determines paths between two points that optimize a cost function of the delay data probability distribution. It can be used to find paths that maximize the probability of reaching a destination within a particular travel deadline. Extensions of the algorithm are further studied for multi-agent path planning. In addition, our novel pre-processing algorithms enable order of magnitudes faster query performance over the state of the art with a proven approximation guarantee.

Our second set of algorithms suggests a distributed method to find paths for multiple agents by introducing a probabilistic path choice achieving global goals such as the social optimum or user equilibrium. Our algorithms find a good solution that considers the congestion caused by each drivers path choice. Although each drivers contribution to congestion may be small, the aggregate is large. This approach, which shows that the global goals can be achieved by local processing using only local information, can be parallelized and sped-up using massive parallel processing.

The suite of algorithms have been implemented as a Web-based software system that enables users to query their paths. Experimental results based on the real data show significant improvement over the existing route planning systems.

City-scale Transportation System

real taxi traffic flow 

real taxi traffic flow

greedy taxi traffic flow 

greedy taxi traffic flow

optimal taxi traffic flow 

optimal taxi traffic flow

We have implemented these algorithms using data from a large set of taxi probes and loop detectors and demonstrate that the resulting traffic routing system can be used for guiding human-driven or autonomous cars. Our fast computation techniques and scalable design enable planning for the larger systems in which our algorithms will function.

Using the developed algorithms and data-driven congestion models, we have implemented a traffic routing system for multiple agents. We have evaluated our algorithm using the sensor data. We have demonstrated that I can achieve the users’ goals while reducing the travel time for the users and for the society when our algorithms are applied. By running very large scale experiments using the real data from 16,000 taxis and 10,000 loop detectors, We have shown that the city-scale congestion can be mitigated by planning drivers’ routes, while incorporating the congestion effects generated by their route choices. When our algorithm is used to a taxi company with 5000 taxis the simulation using the real world taxi traces shows that the company saves two years’ worth of driving in a day's operation. We can imagine how big impact our algorithm can bring to the society if deployed in a city scale.


The fuel economy can be improved by using the data coming from the car. OBD-II terminals can be used to get the various data from car.


Multi-agent control

Our research focuses on multi-agent/robot systems and spans on the fields of robotics, optimization, transportation, and data mining. We are interested in understanding the various types of interactions and dynamics in the multi-agent systems and using it to build more efficient and effective systems. Our research contributes to this endeavor by focusing on the algorithmic and computational issues involved in the study of multi-agent systems.

Specifically, we design practical control, estimation, and learning algorithms for large groups of agents to accomplish a global goal in a decentralized manner. We provide guarantees on the performance and quality of the solution of our algorithms using analytical tools from control theory, optimization, and statistical inference. We use real sensor data to validate these analyses in computational simulations. We implement software systems for multiple agents based on our algorithms and large scale data analysis.

In a long term, we are planning to continue to develop techniques to model and solve various optimization problems in which the agents are partly or completely distributed and can only communicate with their peers. Agents may be self interested or may have different computation/communication capabilities from their peers. We will develop data-driven algorithms that guarantee the quality of the solution dealing with uncertain or unreliable input data. In many applications, solving complex optimization problems while leaking as little private information as possible is important. How to deal with privacy issues while disseminating relevant information across different agents is another research interest.

Congestion pricing

The avenues for future work include multi-agent routing considering spatio-temporal interaction between cars and congestion-aware dynamic road pricing as an on-line control alternative for congestion. We will tackle challenging problems in computation, communication, and control to achieve efficiency and safety in large-scale transportation.