Abstract

Traffic congestion has large economic and social costs. The introduction of autonomous vehicles can potentially reduce this congestion, both by increasing network throughput and by enabling a social planner to incentivize users of autonomous vehicles to take longer routes that can alleviate congestion on more direct roads. We formalize the effects of altruistic autonomy on roads shared between human drivers and autonomous vehicles. In this work, we develop a formal model of road congestion on shared roads based on the fundamental diagram of traffic. We consider a network of parallel roads and provide algorithms that compute optimal equilibria that are robust to additional unforeseen demand. We further plan for optimal routings when users have varying degrees of altruism. We find that even with arbitrarily small altruism, total latency can be unboundedly better than without altruism, and that the best selfish equilibrium can be unboundedly better than the worst selfish equilibrium. We validate our theoretical results through microscopic traffic simulations and show average latency decrease of a factor of 4 from worst-case selfish equilibrium to the optimal equilibrium when autonomous vehicles are altruistic.

Abstract

Modern vehicles have electrical architectures whose complexity grows year after year due to feature growth corresponding to customer expectations. The latest of the expectations, automation of the dynamic driving task however, is poised to bring about some of the largest changes seen so far. In one fell swoop, not only does required functionality for automated driving drastically increase the system complexity, it also removes the fall-back of the human driver who is usually relied upon to handle unanticipated failures after the fact. The need to architect thus requires a greater rigour than ever before, to maintain the level of safety that has been associated with the automotive industry. The work that is part of this thesis has been conducted, in close collaboration with our industrial partner Scania CV AB, within the Vinnova FFI funded project ARCHER. This thesis aims to provide a methodology for architecting during the concept phase of development, using industrial practices and principles including those from safety standards such as ISO 26262. The main contributions of the thesis are in two areas. The first area i.e. Part A contributes, (i) an analysis of the challenges of architecting automated driving, and serves as a motivation for the approach taken in the rest of this thesis, i.e. Part B where the contributions include, (ii) a definition of a viewpoint for functional safety according to the definitions of ISO 42010, (iii) a method to systematically extract information from legacy components and (iv) a process to use legacy information and architect in the presence of uncertainty to provide a work product, the Preliminary Architectural Assumptions (PAA), as required by ISO 26262. The contributions of Part B together comprise a methodology to architect the PAA.   A significant challenge in working with the industry is finding the right fit between idealized principles and practical utility. The methodology in Part B has been judged fit for purpose by different parts of the organization at Scania and multiple case studies have been conducted to assess its usefulness in collaboration with senior architects. The methodology was found to be conducive in both, generating the PAA of a quality that was deemed suitable to the organization and, to find inadequacies in the architecture that had not been found earlier using the previous non-systematic methods. The benefits have led to a commissioning of a prototype tool to support the methodology that has begun to be used in projects related to automation at Scania. The methodology will be refined as the projects progress towards completion using the experiences gained. A further impact of the work is seen in two patent filings that have originated from work on the case studies in Part B. Emanating from needs discovered during the application of the methods, these filed patents (with no prior publications) outline the future directions of research into reference architectures augmented with safety policies, that are safe in the presence of detectable faults and failures. To aid verification of these ideas, work has begun on identifying critical scenarios and their elements in automated driving, and a flexible simulation platform is being designed and developed at KTH to test the chosen critical scenarios. Efterfrågan på nya funktioner leder till en ständigt ökande komplexitet i moderna fordon, speciellt i de inbyggda datorsystemen. Införande av autonoma fordon utgör inte bara det mest aktuella exemplet på detta, utan medför också en av de största förändringar som fordonsbranschen sett. Föraren, som ”back-up” för att hantera oväntade situationer och fel, finns inte längre där vid höggradig automation, och motsvarande funktioner måste realiseras i de inbyggda system vilket ger en drastisk komplexitetsökning. Detta ställer systemarkitekter för stora utmaningar för att se till att nuvarande nivå av funktionssäkerhet bibehålls. Detta forskningsarbete har utförts i nära samarbete med Scania CV AB i det Vinnova (FFI)-finansierade projektet ARCHER. Denna licentiatavhandling har som mål att ta fram en metodik för konceptutveckling av arkitekturer, förankrat i industriell praxis och principer, omfattande bl.a. de som beskrivs i funktionssäkerhetsstandards som ISO 26262. Avhandlingen presenterar resultat inom två områden. Det första området, del A, redovisar, (i) en analys av utmaningar inom arkitekturutveckling för autonoma fordon, vilket också ger en motivering för resterande del av avhandlingen. Det andra området, del B, redovisar, (ii) en definition av en ”perspektivmodell” (en s.k. ”viewpoint” enligt ISO 42010) för funktionssäkerhet, (iii) en metod för att systematiskt utvinna information från existerande komponenter, och (iv) en process som tar fram en arbetsprodukt för ISO 26262 – Preliminära Arkitektur-Antaganden (PAA). Denna process använder sig av information från existerande komponenter – resultat (iii) och förenklar hantering av avsaknad/osäker information under arkitekturarbetet. Resultaten från del B utgör tillsammans en metodik för att ta fram en PAA. En utmaning i forskning är att finna en balans mellan idealisering och praktisk tillämpbarhet. Metodiken i del B har utvärderats i flertalet industriella fallstudier på Scania i samverkan med seniora arkitekter från industrin, och har av dessa bedömts som relevant och praktiskt tillämpningsbar. Erfarenheterna visar att metodiken stödjer framtagandet av PAA’s av   lämplig kvalitet och ger ett systematiskt sätt att hantera osäkerhet under arkitekturutvecklingen. Specifikt så gjorde metoden det möjligt att identifiera komponent-felmoder där arkitekturen inte var tillräcklig för åstadkomma önskad riskreducering, begränsningar som inte hade upptäckts med tidigare metoder. Ett prototypverktyg för att stödja metodiken har utvecklats och börjat användas på Scania i projekt relaterade till autonoma fordon. Metodiken kommer sannolikt att kunna förfinas ytterligare när dessa projekt går mot sitt slut och mer erfarenheter finns tillgängliga. Arbetet i del B har vidare lett till två patentansökningar avseende koncept som framkommit genom fallstudierna. Dessa koncept relaterar till referensarkitekturer som utökats med policies för personsäkerhet (Eng. ”safety”) för att hantera detekterbara felfall, och pekar ut en riktning för framtida forskning. För att stödja verifiering av dessa koncept har arbete inletts för att identifiera kritiska scenarios för autonom körning. En flexibel simuleringsplattform håller också på att designas för att kunna testa kritiska scenarios. Vinnova-FFI funded Project ARCHER

Abstract

Many driver assistance systems such as Adaptive Cruise Control require the identification of the closest vehicle that is in the host vehicle's path. This entails an assignment of detected vehicles to the host vehicle path or neighboring paths. After reviewing approaches to the estimation of the host vehicle path and lane assignment techniques we introduce two methods that are motivated by the rationale to filter measured data as late in the processing stages as possible in order to avoid delays and other artifacts of intermediate filters. These filters generate discrete posterior probability distributions from which a path or "lane" index is extracted by a median estimator. The relative performance of those methods is illustrated by a ROC using experimental data and labeled ground truth data. Comment: 6 pages, 9 figure

Abstract

Recent progress in the field of robotic manipulation has generated interest in fully automatic object packing in warehouses. This paper proposes a formulation of the packing problem that is tailored to the automated warehousing domain. Besides minimizing waste space inside a container, the problem requires stability of the object pile during packing and the feasibility of the robot motion executing the placement plans. To address this problem, a set of constraints are formulated, and a constructive packing pipeline is proposed to solve these constraints. The pipeline is able to pack geometrically complex, non-convex objects while satisfying stability and robot packability constraints. In particular, a new 3D positioning heuristic called Heightmap-Minimization heuristic is proposed, and heightmaps are used to speed up the search. Experimental evaluation of the method is conducted with a realistic physical simulator on a dataset of scanned real-world items, demonstrating stable and high-quality packing plans compared with other 3D packing methods.

Abstract

dvances in sensor technologies, object detection algorithms, planning frameworks and hardware designs have motivated the deployment of robots in warehouse automation. A variety of such applications, like order fulfillment or packing tasks, require picking objects from unstructured piles and carefully arranging them in bins or containers. Desirable solutions need to be low-cost, easily deployable and controllable, making minimalistic hardware choices desirable. The challenge in designing an effective solution to this problem relates to appropriately integrating multiple components, so as to achieve a robust pipeline that minimizes failure conditions. The current work proposes a complete pipeline for solving such packing tasks, given access only to RGB-D data and a single robot arm with a vacuum-based end-effector, which is also used as a pushing finger. To achieve the desired level of robustness, three key manipulation primitives are identified, which take advantage of the environment and simple operations to successfully pack multiple cubic objects. The overall approach is demonstrated to be robust to execution and perception errors. The impact of each manipulation primitive is evaluated by considering different versions of the proposed pipeline, which incrementally introduce reasoning about object poses and corrective manipulation actions.

Abstract

In this paper, we address the problem of time-optimal coordination of mobile robots under kinodynamic constraints along specified paths. We propose a novel approach based on time discretization that leads to a mixed-integer linear programming (MILP) formulation. This problem can be solved using general-purpose MILP solvers in a reasonable time, resulting in a resolution-optimal solution. Moreover, unlike previous work found in the literature, our formulation allows an exact linear modeling (up to the discretization resolution) of second-order dynamic constraints. Extensive simulations are performed to demonstrate the effectiveness of our approach. Comment: Published in 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Abstract

Light Detection and Ranging (LIDAR) sensors play an important role in the perception stack of autonomous robots, supplying mapping and localization pipelines with depth measurements of the environment. While their accuracy outperforms other types of depth sensors, such as stereo or time-of-flight cameras, the accurate modeling of LIDAR sensors requires laborious manual calibration that typically does not take into account the interaction of laser light with different surface types, incidence angles and other phenomena that significantly influence measurements. In this work, we introduce a physically plausible model of a 2D continuous-wave LIDAR that accounts for the surface-light interactions and simulates the measurement process in the Hokuyo URG-04LX LIDAR. Through automatic differentiation, we employ gradient-based optimization to estimate model parameters from real sensor measurements. Comment: Published at ICRA 2020

Abstract

robust estimation of road course and traffic lanes is an essential part of environment perception for next generations of Advanced Driver Assistance Systems and development of self-driving vehicles. In this paper, a flexible method for modeling multiple lanes in a vehicle in real time is presented. Information about traffic lanes, derived by cameras and other environmental sensors, that is represented as features, serves as input for an iterative expectation-maximization method to estimate a lane model. The generic and modular concept of the approach allows to freely choose the mathematical functions for the geometrical description of lanes. In addition to the current measurement data, the previously estimated result as well as additional constraints to reflect parallelism and continuity of traffic lanes, are considered in the optimization process. As evaluation of the lane estimation method, its performance is showcased using cubic splines for the geometric representation of lanes in simulated scenarios and measurements recorded using a development vehicle. In a comparison to ground truth data, robustness and precision of the lanes estimated up to a distance of 120 m are demonstrated. As a part of the environmental modeling, the presented method can be utilized for longitudinal and lateral control of autonomous vehicles.

Abstract

The trend towards autonomous driving and the continuous research in the automotive area, like Advanced Driver Assistance Systems (ADAS), requires an accurate localization under all circumstances. An accurate estimation of the vehicle state is a basic requirement for any trajectory-planning algorithm. Still, even when the introduction of the GPS L5 band promises lane-accuracy, coverage limitations in roofed areas still have to be addressed. In this work, a method for high precision indoor positioning using a LiDAR is presented. The method is based on the combination of motion models with LiDAR measurements, and uses infrastructural elements as positioning references. This allows to estimate the orientation, velocity over ground and position of a vehicle in a Local Tangent Plane (LTP) reference frame. When the outputs of the proposed method are compared to those of an Automotive Dynamic Motion Analyzer (ADMA), mean errors of 1 degree, 0.1 m/s and of 4.7 cm respectively are obtained. The method can be implemented by using a LiDAR sensor as a stand-alone unit. A median runtime of 40.77 us on an Intel i7-6820HQ CPU signals the possibility of real-time processing.

Abstract

We present a sparse representation of model uncertainty for Deep Neural Networks (DNNs) where the parameter posterior is approximated with an inverse formulation of the Multivariate Normal Distribution (MND), also known as the information form. The key insight of our work is that the information matrix, i.e. the inverse of the covariance matrix tends to be sparse in its spectrum. Therefore, dimensionality reduction techniques such as low rank approximations (LRA) can be effectively exploited. To achieve this, we develop a novel sparsification algorithm and derive a cost-effective analytical sampler. As a result, we show that the information form can be scalably applied to represent model uncertainty in DNNs. Our exhaustive theoretical analysis and empirical evaluations on various benchmarks show the competitiveness of our approach over the current methods. Comment: Accepted to the Thirty-seventh International Conference on Machine Learning (ICML) 2020