Vishnu R. Desaraju

Driver Behavior Classification at Intersections and Validation on Large Naturalistic Data Set

The ability to classify driver behavior lays the foundation for more advanced driver assistance systems. In particular, improving safety at intersections has been identified as a high priority due to the large number of intersection-related fatalities. This paper focuses on developing algorithms for estimating driver behavior at road intersections and validating them on real traffic data. It introduces two classes of algorithms that can classify drivers as compliant or violating. They are based on (1) support vector machines and (2) hidden Markov models, which are two very popular machine learning approaches that have been used successfully for classification in multiple disciplines. However, existing work has not explored the benefits of applying these techniques to the problem of driver behavior classification at intersections. The developed algorithms are successfully validated using naturalistic intersection data collected in Christiansburg, VA, through the U.S. Department of Transportation Cooperative Intersection Collision Avoidance System for Violations initiative. Their performances are also compared with those of three traditional methods, and the results show significant improvements with the new algorithms.

Partial order techniques for vehicle collision avoidance: Application to an autonomous roundabout test-bed

In this paper, we employ partial order techniques to develop linear complexity algorithms for guaranteed collision avoidance between vehicles at highway and roundabout mergings. These techniques can be employed by virtue of the rich structure offered by such traffic systems, which constrain vehicles to advance unidirectionally along a path. The algorithms are safe by construction while maintaining the liveness of the system. The proposed algorithms are on-line implemented in a decentralized fashion on an experimental testbed composed of two in-scale communicating vehicles continuously running on an autonomous roundabout system.

Vision-based Landing Site Evaluation and Trajectory Generation Toward Rooftop Landing

Autonomous landing is an essential function for micro air vehicles (MAVs) for many scenarios. We pursue an active perception strategy that enables MAVs with limited onboard sensing and processing capabilities to concurrently assess feasible rooftop landing sites with a vision-based perception system while generating trajectories that balance continued landing site assessment and the requirement to provide visual monitoring of an interest point. The contributions of the work are twofold: (1) a perception system that employs a dense motion stereo approach that determines the 3D model of the captured scene without the need of geo-referenced images, scene geometry constraints, or external navigation aids; and (2) an online trajectory generation approach that balances the need to concurrently explore available rooftop vantages of an interest point while ensuring confidence in the landing site suitability by considering the impact of landing site uncertainty as assessed by the perception system. Simulation and experimental evaluation of the performance of the perception and trajectory generation methodologies are analyzed independently and jointly in order to establish the efficacy of the proposed approach.

Fast nonlinear model predictive control via partial enumeration

In this work, we consider the problem of fast, accurate control of a robot with constrained dynamics. We present a new nonlinear model predictive control (MPC) technique, Nonlinear Partial Enumeration (NPE), that combines online and offline computation in a nonlinear version of the partial enumeration method for MPC, thereby dramatically decreasing the compute time per control iteration. We apply NPE to the problem of MAV flight and demonstrate through a set of simulation trials that NPE outperforms other fast control methodologies during aggressive motion and enables the system to learn a reusable set of local feedback controllers that enable more efficient operation over time.

Hierarchical adaptive planning in environments with uncertain, spatially-varying disturbance forces

This paper presents a hierarchical planning architecture that generates vehicle trajectories that adapt to uncertain, spatially-varying disturbance forces toward enhanced tracking performance. The disturbance force is modeled as a discrete conditional probability distribution that is updated online by local measurements as the vehicle navigates. A global planner identifies the optimal route to the goal and adapts this route according to a cost metric derived from the belief distribution on the disturbance force. A local planner embeds the belief distribution in the trajectory generation process to compute dynamically feasible trajectories along the global plan that evolve with the belief. Simulation studies analyze and demonstrate the increased trajectory tracking accuracy via the proposed methodology with a single vehicle and the impact of the approach to multiple agents performing collaborative inference toward enhanced collective performance.

Experience-driven Predictive Control with Robust Constraint Satisfaction under Time-Varying State Uncertainty.

We present an extension to Experience-driven Predictive Control (EPC) that leverages a Gaussian belief propagation strategy to compute an uncertainty set bounding the evolution of the system state in the presence of time-varying state uncertainty. This uncertainty set is used to tighten the constraints in the predictive control formulation via a chance constrained approach, thereby providing a probabilistic guarantee of constraint satisfaction. The parameterized form of the controllers produced by EPC coupled with online uncertainty estimates ensures this robust constraint satisfaction property persists even as the system switches controllers and experiences variations in the uncertainty model. We validate the online performance and robust constraint satisfaction of the proposed Robust EPC algorithm through a series of experimental trials with a small quadrotor platform subjected to changes in state estimate quality.

Vision-based landing site evaluation and informed optimal trajectory generation toward autonomous rooftop landing

Autonomous landing is an essential function for micro air vehicles (MAVs) for many scenarios. We pursue an active perception strategy that enables MAVs with limited onboard sensing and processing capabilities to concurrently assess feasible rooftop landing sites with a vision-based perception system while generating trajectories that balance continued landing site assessment and the requirement to provide visual monitoring of an interest point. The contributions of the work are twofold: (1) a perception system that employs a dense motion stereo approach that determines the 3D model of the captured scene without the need of geo-referenced images, scene geometry constraints, or external navigation aids; and (2) an online trajectory generation approach that balances the need to concurrently explore available rooftop vantages of an interest point while ensuring confidence in the landing site suitability by considering the impact of landing site uncertainty as assessed by the perception system. Simulation and experimental evaluation of the performance of the perception and trajectory generation methodologies are analyzed independently and jointly in order to establish the efficacy and robustness of the proposed approach.

Multi-Vehicle Lunar Operations Simulation Using SEXTANT

With increasingly higher resolution maps of planetary bodies becoming available, it now possible to predict operational properties of exploration missions using accurate vehicle path planning tools. This, in turn, could lead to a change in the way we view rover operational simulation during design. To this end, the Massachusetts Institute of Technology, along with Aurora Flight Sciences, has enhanced the Surface Exploration Traverse Analysis and Navigation Tool (SEXTANT). Previously developed as a tool to assist astronauts during extra-vehicular activities, SEXTANT now allows a designer to simultaneously plan the path of multiple rovers, each with different properties, within a single exploration mission. It is also paired with a rover modeling tool which estimates the mass of each of the rovers depending on the payload and operational requirements placed on them. It thus provides a feedback loop between operational conditions and system design. SEXTANT therefore has two major functionalities: (1) it allows for the realistic simulation of vehicle traverses to assist in hardware design and (2) it allows for pre-mission multi-vehicle path planning, which in turn allows the user to understand the operational properties of multi-asset lunar exploration. The paper will detail the capabilities of SEXTANT, including its ability to compute optimal paths, energy expenditure, illumination and communication visibility. It also includes a collision avoidance algorithm which enforces safe spacing between vehicles and an energy profile tool which ensures that solar powered rovers receive enough illumination to complete the traverse without running out of energy. The paper includes two cases studies. The first demonstrates how SEXTANT can be used to inform the design of the rover in the early mission concept phase and understand how the path affects the design. The second demonstrates that the collision avoidance algorithm can help plan the path of multiple rovers prior to a mission, thus reducing the software needs onboard the vehicles and allowing for routes to be pre-planned. This in turn increases the overall number of sites visited during the mission, and hence increases the mission’s science return.

Leveraging experience for robust, adaptive nonlinear MPC on computationally constrained systems with time-varying state uncertainty

This paper presents a robust-adaptive nonlinear model predictive control (MPC) technique that leverages past experiences to achieve tractability on computationally constrained systems. We propose a robust extension of the Experience-driven Predictive Control (EPC) algorithm via a Gaussian belief propagation strategy that computes an uncertainty set, bounding the evolution of the system state in the presence of time-varying state uncertainty. This uncertainty set is used to tighten the constraints in the predictive control formulation via a chance-constrained approach, thereby providing a probabilistic guarantee of constraint satisfaction. The parameterized form of the controllers produced by EPC coupled with online uncertainty estimates ensures that this robust constraint satisfaction property persists, even as the system switches controllers and experiences variations in the uncertainty model. We validate the online performance and robust constraint satisfaction of the proposed Robust EPC algorithm through a series of trials with a simulated ground robot and three experimental platforms: (1) a small quadrotor aerial robot executing aggressive maneuvers in wind with degraded state estimates, (2) a skid-steer ground robot equipped with a laser-based localization system, and (3) a hexarotor aerial robot equipped with a vision-based localization system.

Efficient Prioritization in Explicit Adaptive NMPC through Reachable-Space Search

This paper presents a computationally tractable explicit nonlinear model predictive control (NMPC) strategy that models and adapts to changes in plant dynamics. Explicit NMPC techniques enumerate all controllers derived from an NMPC formulation. However, the resulting database is exponential in the number of constraints and prohibitive for fast, online queries. Therefore, we propose to construct a database that is restricted to operation within the system’s reachable set under NMPC. To identify this reduced controller set, we construct a randomized search tree to explore the set of trajectories within the reachable set and apply the Experience-driven Predictive Control (EPC) algorithm to construct the database incrementally during the search. Additionally, we model transitions between controllers as a Markov chain with transition probabilities that inform a partial ordering on successors for each controller, thus enabling efficient search of the database at runtime. The resulting Explicit EPC algorithm thus consists of two phases: 1) offline generation of a simplified controller database and 2) efficient online application of the stored controllers. A set of simulation studies, including attitude control of a quadrotor micro air vehicle, demonstrate that the proposed approach enables the use of explicit adaptive NMPC for problems that would otherwise yield prohibitively large databases. We also present experimental evaluation of the proposed Explicit EPC algorithm implemented onboard a nano-quadrotor, thereby demonstrating that this approach enables adaptive NMPC on severely compute-constrained platforms.