Optimization Based Verification and Synthesis for Safe Autonomy

161770-Thumbnail Image.png
Description
Autonomous systems should satisfy a set of requirements that guarantee their safety, efficiency, and reliability when working under uncertain circumstances. These requirements can have financial, or legal implications or they can describe what is assigned to autonomous systems.As a result,

Autonomous systems should satisfy a set of requirements that guarantee their safety, efficiency, and reliability when working under uncertain circumstances. These requirements can have financial, or legal implications or they can describe what is assigned to autonomous systems.As a result, the system controller needs to be designed in order to comply with these - potentially complicated - requirements, and the closed-loop system needs to be tested and verified against these requirements. However, when the complexity of the system and its requirements increases, designing a requirement-based controller for the system and analyzing the closed-loop system against the requirement becomes very challenging. In this case, existing design and test methodologies based on trial-and-error would fail, and hence disciplined scientific approaches should be considered. To address some of these challenges, in this dissertation, I present different methods that facilitate efficient testing, and control design based on requirements: 1. Gradient-based methods for improved optimization-based testing, 2. Requirement-based learning for the design of neural-network controllers, 3. Methods based on barrier functions for designing control inputs that ensure the satisfaction of safety constraints.
Date Created
2021
Agent

Physically Realizable Targeted Adversarial Attacks on Autonomous Driving

Description
Autonomous Driving (AD) systems are being researched and developed actively in recent days to solve the task of controlling the vehicles safely without human intervention. One method to solve such task is through deep Reinforcement Learning (RL) approach. In dee

Autonomous Driving (AD) systems are being researched and developed actively in recent days to solve the task of controlling the vehicles safely without human intervention. One method to solve such task is through deep Reinforcement Learning (RL) approach. In deep RL, the main objective is to find an optimal control behavior, often called policy performed by an agent, which is AD system in this case. This policy is usually learned through Deep Neural Networks (DNNs) based on the observations that the agent perceives along with rewards feedback received from environment.However, recent studies demonstrated the vulnerability of such control policies learned through deep RL against adversarial attacks. This raises concerns about the application of such policies to risk-sensitive tasks like AD. Previous adversarial attacks assume that the threats can be broadly realized in two ways: First one is targeted attacks through manipu- lation of the agent’s complete observation in real time and the other is untargeted attacks through manipulation of objects in environment. The former assumes full access to the agent’s observations at almost all time, while the latter has no control over outcomes of attack. This research investigates the feasibility of targeted attacks through physical adver- sarial objects in the environment, a threat that combines the effectiveness and practicality. Through simulations on one of the popular AD systems, it is demonstrated that a fixed optimal policy can be malfunctioned over time by an attacker e.g., performing an unintended self-parking, when an adversarial object is present. The proposed approach is formulated in such a way that the attacker can learn a dynamics of the environment and also utilizes common knowledge of agent’s dynamics to realize the attack. Further, several experiments are conducted to show the effectiveness of the proposed attack on different driving scenarios empirically. Lastly, this work also studies robustness of object location, and trade-off between the attack strength and attack length based on proposed evaluation metrics.
Date Created
2021
Agent

DSL for Spatio-Temporal Perception Logic Specifications

148109-Thumbnail Image.png
Description

System and software verification is a vital component in the development and reliability of cyber-physical systems - especially in critical domains where the margin of error is minimal. In the case of autonomous driving systems (ADS), the vision perception subsystem

System and software verification is a vital component in the development and reliability of cyber-physical systems - especially in critical domains where the margin of error is minimal. In the case of autonomous driving systems (ADS), the vision perception subsystem is a necessity to ensure correct maneuvering of the environment and identification of objects. The challenge posed in perception systems involves verifying the accuracy and rigidity of detections. The use of Spatio-Temporal Perception Logic (STPL) enables the user to express requirements for the perception system to verify, validate, and ensure its behavior; however, a drawback to STPL involves its accessibility. It is limited to individuals with an expert or higher-level knowledge of temporal and spatial logics, and the formal-written requirements become quite verbose with more restrictions imposed. In this thesis, I propose a domain-specific language (DSL) catered to Spatio-Temporal Perception Logic to enable non-expert users the ability to capture requirements for perception subsystems while reducing the necessity to have an experienced background in said logic. The domain-specific language for the Spatio-Temporal Perception Logic is built upon the formal language with two abstractions. The main abstraction captures simple programming statements that are translated to a lower-level STPL expression accepted by the testing monitor. The STPL DSL provides a seamless interface to writing formal expressions while maintaining the power and expressiveness of STPL. These translated equivalent expressions are capable of directing a standard for perception systems to ensure the safety and reduce the risks involved in ill-formed detections.

Date Created
2021-05
Agent

Domain Concretization from Examples: Addressing Missing Domain Knowledge via Robust Planning

158851-Thumbnail Image.png
Description
Most planning agents assume complete knowledge of the domain, which may not be the case in scenarios where certain domain knowledge is missing. This problem could be due to design flaws or arise from domain ramifications or qualifications. In such

Most planning agents assume complete knowledge of the domain, which may not be the case in scenarios where certain domain knowledge is missing. This problem could be due to design flaws or arise from domain ramifications or qualifications. In such cases, planning algorithms could produce highly undesirable behaviors. Planning with incomplete domain knowledge is more challenging than partial observability in the sense that the planning agent is unaware of the existence of such knowledge, in contrast to it being just unobservable or partially observable. That is the difference between known unknowns and unknown unknowns.

In this thesis, I introduce and formulate this as the problem of Domain Concretization, which is inverse to domain abstraction studied extensively before. Furthermore, I present a solution that starts from the incomplete domain model provided to the agent by the designer and uses teacher traces from human users to determine the candidate model set under a minimalistic model assumption. A robust plan is then generated for the maximum probability of success under the set of candidate models. In addition to a standard search formulation in the model-space, I propose a sample-based search method and also an online version of it to improve search time. The solution presented has been evaluated on various International Planning Competition domains where incompleteness was introduced by deleting certain predicates from the complete domain model. The solution is also tested in a robot simulation domain to illustrate its effectiveness in handling incomplete domain knowledge. The results show that the plan generated by the algorithm increases the plan success rate without impacting action cost too much.
Date Created
2020
Agent

Bioman: Discrete-event Simulator to Analyze Operations for Car-T Cell Therapy Manufacturing

158530-Thumbnail Image.png
Description
The success of genetically-modified T-cells in treating hematological malignancies has accelerated the research timeline for Chimeric Antigen Receptor-T (CAR-T) cell therapy. Since there are only two approved products (Kymriah and Yescarta), the process knowledge is limited. This leads to a

The success of genetically-modified T-cells in treating hematological malignancies has accelerated the research timeline for Chimeric Antigen Receptor-T (CAR-T) cell therapy. Since there are only two approved products (Kymriah and Yescarta), the process knowledge is limited. This leads to a low efficiency at manufacturing stage with serious challenges corresponding to high cost and scalability. In addition, the individualized nature of the therapy limits inventory and creates a high risk of product loss due to supply chain failure. The sector needs a new manufacturing paradigm capable of quickly responding to individualized demands while considering complex system dynamics.

The research formulates the problem of Chimeric Antigen Receptor-T (CAR-T) manufacturing design, understanding the performance for large scale production of personalized therapies. The solution looks to develop a simulation environment for bio-manufacturing systems with single-use equipment. The result is BioMan: a discrete-event simulation model that considers the role of therapy's individualized nature, type of processing and quality-management policies on process yield and time, while dealing with the available resource constraints simultaneously. The tool will be useful to understand the impact of varying factor inputs on Chimeric Antigen Receptor-T (CAR-T) cell manufacturing and will eventually facilitate the decision-maker to finalize the right strategies achieving better processing, high resource utilization, and less failure rates.
Date Created
2020
Agent

Self-Organization of Multi-Agent Systems Using Markov Chain Models

158221-Thumbnail Image.png
Description
The problem of modeling and controlling the distribution of a multi-agent system has recently evolved into an interdisciplinary effort. When the agent population is very large, i.e., at least on the order of hundreds of agents, it is important that

The problem of modeling and controlling the distribution of a multi-agent system has recently evolved into an interdisciplinary effort. When the agent population is very large, i.e., at least on the order of hundreds of agents, it is important that techniques for analyzing and controlling the system scale well with the number of agents. One scalable approach to characterizing the behavior of a multi-agent system is possible when the agents' states evolve over time according to a Markov process. In this case, the density of agents over space and time is governed by a set of difference or differential equations known as a {\it mean-field model}, whose parameters determine the stochastic control policies of the individual agents. These models often have the advantage of being easier to analyze than the individual agent dynamics. Mean-field models have been used to describe the behavior of chemical reaction networks, biological collectives such as social insect colonies, and more recently, swarms of robots that, like natural swarms, consist of hundreds or thousands of agents that are individually limited in capability but can coordinate to achieve a particular collective goal.

This dissertation presents a control-theoretic analysis of mean-field models for which the agent dynamics are governed by either a continuous-time Markov chain on an arbitrary state space, or a discrete-time Markov chain on a continuous state space. Three main problems are investigated. First, the problem of stabilization is addressed, that is, the design of transition probabilities/rates of the Markov process (the agent control parameters) that make a target distribution, satisfying certain conditions, invariant. Such a control approach could be used to achieve desired multi-agent distributions for spatial coverage and task allocation. However, the convergence of the multi-agent distribution to the designed equilibrium does not imply the convergence of the individual agents to fixed states. To prevent the agents from continuing to transition between states once the target distribution is reached, and thus potentially waste energy, the second problem addressed within this dissertation is the construction of feedback control laws that prevent agents from transitioning once the equilibrium distribution is reached. The third problem addressed is the computation of optimized transition probabilities/rates that maximize the speed at which the system converges to the target distribution.
Date Created
2020
Agent

Mission and Motion Planning for Multi-robot Systems in Constrained Environments

157990-Thumbnail Image.png
Description
As robots become mechanically more capable, they are going to be more and more integrated into our daily lives. Over time, human’s expectation of what the robot capabilities are is getting higher. Therefore, it can be conjectured that often robots

As robots become mechanically more capable, they are going to be more and more integrated into our daily lives. Over time, human’s expectation of what the robot capabilities are is getting higher. Therefore, it can be conjectured that often robots will not act as human commanders intended them to do. That is, the users of the robots may have a different point of view from the one the robots do.

The first part of this dissertation covers methods that resolve some instances of this mismatch when the mission requirements are expressed in Linear Temporal Logic (LTL) for handling coverage, sequencing, conditions and avoidance. That is, the following general questions are addressed:

* What cause of the given mission is unrealizable?

* Is there any other feasible mission that is close to the given one?

In order to answer these questions, the LTL Revision Problem is applied and it is formulated as a graph search problem. It is shown that in general the problem is NP-Complete. Hence, it is proved that the heuristic algorihtm has 2-approximation bound in some cases. This problem, then, is extended to two different versions: one is for the weighted transition system and another is for the specification under quantitative preference. Next, a follow up question is addressed:

* How can an LTL specified mission be scaled up to multiple robots operating in confined environments?

The Cooperative Multi-agent Planning Problem is addressed by borrowing a technique from cooperative pathfinding problems in discrete grid environments. Since centralized planning for multi-robot systems is computationally challenging and easily results in state space explosion, a distributed planning approach is provided through agent coupling and de-coupling.

In addition, in order to make such robot missions work in the real world, robots should take actions in the continuous physical world. Hence, in the second part of this thesis, the resulting motion planning problems is addressed for non-holonomic robots.

That is, it is devoted to autonomous vehicles’ motion planning in challenging environments such as rural, semi-structured roads. This planning problem is solved with an on-the-fly hierarchical approach, using a pre-computed lattice planner. It is also proved that the proposed algorithm guarantees resolution-completeness in such demanding environments. Finally, possible extensions are discussed.
Date Created
2019
Agent

Sample-Efficient Reinforcement Learning of Robot Control Policies in the Real World

157799-Thumbnail Image.png
Description
The goal of reinforcement learning is to enable systems to autonomously solve tasks in the real world, even in the absence of prior data. To succeed in such situations, reinforcement learning algorithms collect new experience through interactions with the environment

The goal of reinforcement learning is to enable systems to autonomously solve tasks in the real world, even in the absence of prior data. To succeed in such situations, reinforcement learning algorithms collect new experience through interactions with the environment to further the learning process. The behaviour is optimized by maximizing a reward function, which assigns high numerical values to desired behaviours. Especially in robotics, such interactions with the environment are expensive in terms of the required execution time, human involvement, and mechanical degradation of the system itself. Therefore, this thesis aims to introduce sample-efficient reinforcement learning methods which are applicable to real-world settings and control tasks such as bimanual manipulation and locomotion. Sample efficiency is achieved through directed exploration, either by using dimensionality reduction or trajectory optimization methods. Finally, it is demonstrated how data-efficient reinforcement learning methods can be used to optimize the behaviour and morphology of robots at the same time.
Date Created
2019
Agent

Activity Specification for Time-based Discrete Event Simulation Models

157772-Thumbnail Image.png
Description
Computational models for relatively complex systems are subject to many difficulties, among which is the ability for the models to be discretely understandable and applicable to specific problem types and their solutions. This demands the specification of a dynamic system

Computational models for relatively complex systems are subject to many difficulties, among which is the ability for the models to be discretely understandable and applicable to specific problem types and their solutions. This demands the specification of a dynamic system as a collection of models, including metamodels. In this context, new modeling approaches and tools can help provide a richer understanding and, therefore, the development of sophisticated behavior in system dynamics. From this vantage point, an activity specification is proposed as a modeling approach based on a time-based discrete event system abstraction. Such models are founded upon set-theoretic principles and methods for modeling and simulation with the intent of making them subject to specific and profound questions for user-defined experiments.

Because developing models is becoming more time-consuming and expensive, some research has focused on the acquisition of concrete means targeted at the early stages of component-based system analysis and design. The model-driven architecture (MDA) framework provides some means for the behavioral modeling of discrete systems. The development of models can benefit from simplifications and elaborations enabled by the MDA meta-layers, which is essential for managing model complexity. Although metamodels pose difficulties, especially for developing complex behavior, as opposed to structure, they are advantageous and complementary to formal models and concrete implementations in programming languages.

The developed approach is focused on action and control concepts across the MDA meta-layers and is proposed for the parallel Discrete Event System Specification (P-DEVS) formalism. The Unified Modeling Language (UML) activity meta-models are used with syntax and semantics that conform to the DEVS formalism and its execution protocol. The notions of the DEVS component and state are used together according to their underlying system-theoretic foundation. A prototype tool supporting activity modeling was developed to demonstrate the degree to which action-based behavior can be modeled using the MDA and DEVS. The parallel DEVS, as a formal approach, supports identifying the semantics of the UML activities. Another prototype was developed to create activity models and support their execution with the DEVS-Suite simulator, and a set of prototypical multiprocessor architecture model specifications were designed, simulated, and analyzed.
Date Created
2019
Agent

Search-based Test Generation for Automated Driving Systems: From Perception to Control Logic

157060-Thumbnail Image.png
Description
Automated driving systems are in an intensive research and development stage, and the companies developing these systems are targeting to deploy them on public roads in a very near future. Guaranteeing safe operation of these systems is crucial as they

Automated driving systems are in an intensive research and development stage, and the companies developing these systems are targeting to deploy them on public roads in a very near future. Guaranteeing safe operation of these systems is crucial as they are planned to carry passengers and share the road with other vehicles and pedestrians. Yet, there is no agreed-upon approach on how and in what detail those systems should be tested. Different organizations have different testing approaches, and one common approach is to combine simulation-based testing with real-world driving.

One of the expectations from fully-automated vehicles is never to cause an accident. However, an automated vehicle may not be able to avoid all collisions, e.g., the collisions caused by other road occupants. Hence, it is important for the system designers to understand the boundary case scenarios where an autonomous vehicle can no longer avoid a collision. Besides safety, there are other expectations from automated vehicles such as comfortable driving and minimal fuel consumption. All safety and functional expectations from an automated driving system should be captured with a set of system requirements. It is challenging to create requirements that are unambiguous and usable for the design, testing, and evaluation of automated driving systems. Another challenge is to define useful metrics for assessing the testing quality because in general, it is impossible to test every possible scenario.

The goal of this dissertation is to formalize the theory for testing automated vehicles. Various methods for automatic test generation for automated-driving systems in simulation environments are presented and compared. The contributions presented in this dissertation include (i) new metrics that can be used to discover the boundary cases between safe and unsafe driving conditions, (ii) a new approach that combines combinatorial testing and optimization-guided test generation methods, (iii) approaches that utilize global optimization methods and random exploration to generate critical vehicle and pedestrian trajectories for testing purposes, (iv) a publicly-available simulation-based automated vehicle testing framework that enables application of the existing testing approaches in the literature, including the new approaches presented in this dissertation.
Date Created
2019
Agent