In this article, we address the problem of motion planning and control at the limits of handling, under locally varying traction conditions. We propose a novel solution method where traction variations over the prediction horizon are represented by time-varying tire force constraints, derived from a predictive friction estimate. A \CFTOClong (\CFTOCshort) is solved in a receding horizon fashion, imposing these time-varying constraints. Furthermore, our method features an integrated sampling augmentation procedure that addresses the problems of infeasibility and sensitivity to local minima that arise at abrupt constraint alterations, for example, due to sudden friction changes. We validate the proposed algorithm on a Volvo FH16 heavy-duty vehicle, in a range of critical scenarios. Experimental results indicate that traction adaptive motion planning and control improves the vehicle's capacity to avoid accidents, both when adapting to low local traction, by ensuring dynamic feasibility of the planned motion, and when adapting to high local traction, by realizing high traction utilization. 

The deployment of autonomous vehicles on public roads calls for the development of methods that are reliably able to mitigate injury severity in case of unavoidable collisions. This study proposes a data-driven motion planning method capable of minimizing injury severity for vehicle occupants in unavoidable collisions. The method is based on establishing a metric that models the relationship between impact location and injury severity using real accident data, and subsequently including it in the cost function of a motion planning framework. The vehicle dynamics and associated constraints are considered through a precomputed trajectory library, which is generated by solving an optimal control problem. This allows for efficient computation as well as an accurate representation of the vehicle. The proposed motion planning approach is evaluated by simulation, and it is shown that the trajectory associated with the minimum cost mitigates the collision severity for occupants of passenger vehicles involved in the collision.

Traction adaptive motion planning and control has the potential to improve the ability of automated vehicles to avoid accidents in critical situations. However, such functionality requires an accurate friction estimate for the road ahead of the vehicle, that is updated in real time. Current state of the art friction estimation techniques include high accuracy local friction estimation in the presence of tire slip, as well as rough classification of the road surface ahead of the vehicle, based on forward looking camera. In this paper, we show that neither of these techniques in isolation yield satisfactory behavior when deployed with traction adaptive motion planning and control functionality. However, we also identify that the two techniques are complementary in terms of the key properties accuracy, availability and foresight, indicating that combining the techniques may be meaningful. To this end, we propose a fusion method based on heteroscedastic gaussian process regression, and present initial simulation based results indicating that near-optimal traction adaptive motion planning and control can be achieved by such a method.

The road traffic environment is inherently uncertain and unpredictable. An automated vehicle (AV) deployed in such an environment will eventually experience unforeseen critical situations, i.e., situations in which the probability of having an accident is rapidly increased compared to a nominal driving situation. Critical situations can occur for example due to internal faults or performance limitations of the AV, abrupt changes in operational conditions or unexpected behavior from other road users. In such critical situations, the first priority for vehicle motion control is to reduce the risk of imminent accident. If needed, the full physical capacity of the vehicle should be employed to accomplish this. These unique circumstances distinguish automated driving in critical situations from the nominal case. 

This work aims to tackle the problem of motion planning and control in such critical situations. We determine a set of characteristics that signify the motion planning and control problem in critical situations, in relation to state of the art algorithms. Further, we incrementally develop a motion planning and control framework, tailored for the particular circumstances of critical situations. In its current form, the framework uses a combination of numerical optimization, trajectory rollout and constraint adaptation, to allow motion planning and control with respect to time-varying actuation capabilities, while realizing a range of behaviors to mitigate accident risk in a range of critical situations. 

Results for the research work are generated by exposing the framework to several categories of critical situations in a combination of simulations and full scale vehicle tests. We present the following main findings: (1) Inclusion of risk levels of stopping locations at the local planning level generates satisfactory motion behavior in the evaluated critical situations, enabling a combined assessment of risk of the maneuver and of the stopping location. (2) Traction adaptive motion planning and control improves the capacity of autonomous vehicles to reduce accident risk in critical situations, both when adapting to deteriorated and when adapting to improved traction in a range of tested critical situations. (3) State of the art friction estimation algorithms are insufficient for traction adaptive motion planning in terms of combined requirements on accuracy, availability and foresight. However, fusion of multiple estimation paradigms show potential to yield near-optimal performance. 

The combined contributions of this thesis are intended as a step towards further improving accident avoidance performance of automated vehicles and driver assistance systems in critical situations. However, much research work remains to be done in this field. We emphasize the need for further research efforts in terms of experimentally evaluating the impact of motion planning and control concepts on accident avoidance performance in critical situations. 

Vägtrafikmiljön är oförutsägbar. Autonoma vägfordon i en sådan miljö kommer tids nog att hamna i oförutsedda kritiska situationer, det vill säga situationer där risken för en trafikolycka är markant högre än vid nominell körning. Kritiska situationer kan orsakas av exempelvis interna fel eller prestandabegränsningar hos autonomisystemet, av plötsliga förändringar i operationella förhållanden eller av oförutsett agerande hos medtrafikanter. I kritiska situationer är passagerarkomfort inte längre en prioritet, utan fordonets fullständiga manöverförmåga kan utnyttjas för att minimera olycksrisken. Dessa omständigheter skiljer autonom körning i kritiska situationer från det nominella fallet.  

Forskningsinriktningen för denna avhandling är rörelseplanering och styrning av autonoma fordon i kritiska situationer. Vi presenterar en uppsättning egenskaper som kännetecknar detta specifika problem, i relation till ledande algoritmer för rörelseplanering och styrning. Vi presenterar också vår egen stegvis utvecklade metod för att angripa problemet. I sin nuvarande form består metoden av en kombination av optimeringsbaserad och samplingsbaserad trajektorieplanering med tidsvarierande dynamik och bivillkor. Metoden gör det möjligt att representera tidsvarierande dynamik och dynamiska begränsningar hos fordonet (till exempel till följd av varierande vägförhållanden) vid planering av en mängd olika manövertyper som kan minska olycksrisken i kritiska situationer. 

Resultaten i forskningsarbetet har genererats genom att testa metoden i ett flertal typer av kritiska situationer som har iscensatts genom en kombination av simuleringsmiljöer och experiment med fullskaliga autonoma testfordon. De huvudsakliga slutsatserna från forskningsarbetet är följande: (1) Att inkludera risknivån hos alternativa stoppositioner på den lokala planeringsnivån genererar tillfredsställande rörelsebeteende vid exempelvis interna fel hos autonomisystemet. Detta möjliggör en sammantagen riskbedömning för manöver och stopposition. (2) Väglagsanpassad rörelseplanering och styrning förbättrar autonoma fordons förmåga att reducera olycksrisk i kritiska situationer, både vid anpassning till försämrade och till förbättrade vägförhållanden. (3) Ledande metoder för skattning av vägfriktion har inte tillfredställande prestanda för väglagsanpassad rörelseplanering och styrning med avseende på kombinerade krav på precision, tillgänglighet och framsynthet, när de används var för sig. Dock är det möjligt att kombinera estimat från olika sensorslag till ett friktionsestimat som ger närmast optimalt rörelsebeteende då det används i kombination med väglagsanpassad rörelseplanering och styrning.

Vår förhoppning är att de sammanlagda forskningsbidragen från denna avhandling kan komma att bidra till fortsatta prestandaförbättringar hos system avsedda att minska olycksrisken i kritiska situationer, både för autonoma fordon och för förarstödsystem. Det finns dock mycket kvar att göra inom detta forskningsfält. Vi vill särskilt framhäva behovet av ytterligare forskningsinitiativ rörande experimentell utvärdering av nya koncept för rörelseplanering och styrning, med avseende på förmågan att minska olycksrisken i kritiska situationer.

An articulated vehicle is a two-body design capable of precise maneuvering around obstacles, while carrying heavy loads over rough terrain. In the context of path planning for automated articulated vehicles, it is desirable to fully utilize the maneuverability of the vehicle to enable autonomous operation in confined areas. In this paper we study the impact of model accuracy in an optimization based path planner for an articulated vehicle. For this purpose, we compare the traditional kinematic bicycle model with a two-body articulated model. We evaluate performance in terms of path length, path quality, success rate and computation time through performing test queries in artificial environments and through experiments on a full scale articulated hauler. Results show that for simple, unidirectional maneuvers, performance differences are small, but for more difficult bidirectional maneuvers, the articulated model produces shorter and higher quality paths at a higher success rate. However, the articulated model has 2.75 times longer computation time on average.

In this paper, we tackle the problem of trajectory planning and control of a vehicle under locally varying traction limitations, in the presence of suddenly appearing obstacles. We employ concepts from adaptive model predictive control for run-time adaptation of tire force constraints that are imposed by local traction conditions. To solve the resulting optimization problem for real-time control synthesis with such time varying constraints, we propose a novel numerical scheme based on Real Time Iteration Sequential Quadratic Programming (RTI-SQP), which we call Sampling Augmented Adaptive RTI (SAA-RTI). Sampling augmentation of conventional RTI-SQP provides additional feasible candidate trajectories for warm-starting the optimization procedure. Thus, the proposed SAA-RTI algorithm enables real time constraint adaptation and reduces sensitivity to local minima. Through extensive numerical simulations we demonstrate that our method increases the vehicle's capacity to avoid accidents in scenarios with unanticipated obstacles and locally varying traction, compared to equivalent non-adaptive control schemes and traditional planning and tracking approaches. 

Autonomous vehicles apply pertinent planning and control algorithms under different driving conditions. The mode switch between these algorithms should also be autonomous. On top of the nominal planners, a safe fallback routine is needed to stop the vehicle at a safe position if nominal operational conditions are violated, such as for a system failure. This paper describes the design and formal verification of a supervisor to manage all requirements for mode switching between nominal planners, and additional requirements for switching to a safe stop trajectory planner that acts as the fallback routine. The supervisor is designed via a model-based approach and its abstraction is formally verified by model checking. The supervisor is implemented and integrated with the Research Concept Vehicle, an experimental research and demonstration vehicle developed at the KTH Royal Institute of Technology. Simulations and experiments show that the vehicle is able to autonomously drive in a safe manner between two parking lots and can successfully come to a safe stop upon GPS sensor failure.

This paper addresses the control of a highly automated vehicle in a traffic scenario, where colliding with other traffic agents is unavoidable. Such a critical situation could be the result of a fault in the vehicle, late obstacle detection or the presence of an aggressive driver. We provide an approach that allows the vehicle’s control system to choose the manoeuvre that is likely to lead to the least severe injuries to vehicle occupants.The approach involves the off-line solving of an optimal control problem to create a set of trajectories based on controlling the steering angle rate and the braking rate at the vehicle’s limits. Occupant injury severity prediction, based on accident data with the focus on impact location, is used by a real-time collision control algorithm to choose a trajectory from the pre-computed optimal set. A simulation set-up is presented to illustrate the idea of the collision control algorithm in a simple scenario involving dynamic traffic agents.

The complexity of automated driving poses challenges for providing safety assurance. Focusing on the architecting of an Autonomous Driving Intelligence (ADI), i.e. the computational intelligence, sensors and communication needed for high levels of automated driving, we investigate so called safety supervisors that complement the nominal functionality. We present a problem formulation and a functional architecture of a fault-tolerant ADI that encompasses a nominal and a safety supervisor channel. We then discuss the sources of hazardous events, the division of responsibilities among the channels, and when the supervisor should take over. We conclude with identified directions for further work.

Cooperative automated driving is a promising development in reducing energy consumption and emissions, increasing road safety, and improving traffic flow. The Grand Cooperative Driving Challenge (GCDC) 2016 was an implementation oriented project with the aim to accelerate research and development in the field. This paper describes the development of the two vehicle systems with which KTH participated in GCDC 2016. It presents a reference system architecture for collaborative automated driving as well as its instantiation on two conceptually different vehicles: a Scania truck and the research concept vehicle, built at KTH. We describe the common system architecture, as well as the implementation of a selection of shared and individual system functionalities, such as V2X communication, localization, state estimation, and longitudinal and lateral control. We also present a novel approach to trajectory tracking control for a four-wheel steering vehicle using model predictive control and a novel method for achieving fair data age distribution in vehicular communications.