Major Ongoing Projects

Intelligent Motion Planning and Control for Autonomous Vehicles

Autonomous vehicles – aircraft, cars, rovers, over- and underwater vehicles that can move in the real world by themselves without human pilotage – have gained immense importance not only due to the broad spectrum of their potential military and civilian applications, but also due to the concurrent development of sensor technology and embedded systems that enable the realization of true autonomy. These vehicles may be assigned tasks that are dull and/or repetitive, such as mobile surveillance or cleaning and maintenance; tasks that are dangerous for humans, such as military transportation via hostile territory, large-scale fire-fighting, and repair and recovery operations in chemical plants and nuclear reactors; or tasks that are prohibitively expensive for humans to execute, such as the exploration of celestial bodies.

alt text 

Firstly, we are interested in the optimal motion planning and control problem, which involves finding control inputs (e.g. steering, throttle) that enable the vehicle’s desired motion. Optimal control theory has been studied for over three hundred years, but new computational strategies are required to develop real-time algorithms.

Secondly, we are interested in the intelligent control problem, which involves developing and integrating artificial intelligence (AI) algorithms with control algorithms to enable the vehicle to achieve complex tasks specified in human-like language (e.g. get me to my workplace ASAP, drop my friend close to his workplace, find a cheap parking spot and wait until I need you again; also save gasoline as much as possible).  AI and automatic control have been traditionally disparate academic disciplines; furthermore, AI and control problems are formulated using fundamentally different mathematical tools, which makes their integration challenging.

Our preliminary approach to the solution of these problems is the H-cost motion-planning algorithm. The general idea is to modify standard workspace cell decomposition-based path-planning to accommodate constraints due to vehicle dynamics. To this end, costs on successions of edges in the workspace cell decomposition graph are considered. The image on the left shows a result of implementation of this algorithm for a particle with a constraint on the magnitude of acceleration. Compared to randomized sampling-based planning, this algorithm results in a trajectory with far lower cost, while satisfying workspace constraints (i.e. avoiding obstacles) and vehicle dynamical constraints.

Related publication: IEEE T-Ro, 28(2), p. 379

alt text 

The basic H-cost algorithm is somewhat slow, and scales poorly. To improve computational efficiency, it is implemented with multiresolution cell decompositions (see lower figure to the left). Intuitively, vehicle dynamics must be considered in motion plans for immediate execution; coarse approximations suffice for longer-term plans. This notion is captured by employing the H-cost approach only in the fine resolution window (the upper figure on the left illustrates the execution of the H-cost approach within this window).

Related publication: IEEE T-SMCB, 42(5), p. 1455

Current research includes:

  • motion-planning using the H-cost approach to satisfy temporal logic specifications

  • incremental planning to accommodate real-time computational issues such as bounded execution time

  • abstractions based on the geometric characteristics of vehicle motion

  • motion-planning for unmanned aerial vehicles subject to airframe structural health condition constraints

Experimental Validation with Small Aerial Vehicles

alt text 

We are currently involved in developing experimental platforms involving micro- and mini- helicopters and micro- (fixed wing) aerial vehicles for validation and testing of motion-planning and control algorithms. The current approach is to install sensors and autopilots on commercial off-the-shelf RC aerial vehicles.

Accident Analysis and System Safety of Hierarchical Systems

alt text 

The risk analysis and system safety literature often reports on “organizational” accidents or “system” accidents in sociotechnical systems. Whereas a “system theoretic approach” to accident analysis and safety has recently been advanced in the literature, formal system theoretic concepts of hierarchical and multilevel systems have been absent from the discussions of safety as a system theoretic problem. To address this gap, we introduce the concepts of coordinability and consistency from the analysis of hierarchical and multilevel systems to the risk analysis and system safety community, and we investigate the applicability of these concepts to system safety via an illustrative example.

Current research includes the applications of these concepts for the safety of small UAVs in civilian applications.

Relevant publication: Risk Analysis, 33(3), p. 420

Other Projects

Real-time Robust Flare Planning for Landing in Crosswinds

alt text 

We address the problem of planning flare trajectories – the short phase between the glide slope and touchdown in fixed-wing aircraft landing – subject to wind disturbances. First, we determine a set of polynomial paths as candidate flare paths, and we determine constraints on the polynomial coefficients corresponding to input constraints for an aircraft dynamical model. Next, we determine bounds on the first-order variations of these polynomial coefficients to characterize allowable deviations from the polynomial paths without violating the input constraints. Finally, we verify safe landing via tracking of each of these candidate paths using randomized sampling-based falsification methods.

Randomized Sampling-based Optimal Motion-Planning for the Halfcar Model

alt text 

Real-time motion planning techniques for autonomous terrestrial vehicles traditionally use low-fidelity, low-dimensional vehicle dynamical model. For high-speed driving, such methods are unable to use advantageously the vehicle’s full envelope of maneuvering capabilities. To remedy this problem, we investigate motion planning with the relatively high-fidelity half-car dynamical model. Specifically, we investigate the application of the RRT* optimal motion planning algorithm for this problem, and we develop a computationally efficient local steering algorithm – a crucial component in the RRT* algorithm.

Relevant publication: ACC 2013