Ilge Akkaya

Ilge Akkaya

Member of Technical Staff @OpenAI

Posts by Collection

portfolio

publications

Systems engineering for industrial cyber–physical systems using aspects

Published in Proceedings of the IEEE, 2016

One of the biggest challenges in cyber–physical system (CPS) design is their intrinsic complexity, heterogeneity, and multidisciplinary nature. Emerging distributed CPSs integrate a wide range of heterogeneous aspects such as physical dynamics, control, machine learning, and error handling. Furthermore, system components are often distributed over multiple physical locations, hardware platforms, and communication networks. While model-based design (MBD) has tremendously improved the design process, CPS design remains a difficult task. Models are meant to improve understanding of a system, yet this quality is often lost when models become too complicated. In this paper, we show how to use aspect-oriented (AO) modeling techniques in MBD as a systematic way to segregate domains of expertise and crosscutting concerns within the model. We demonstrate these concepts on actor-oriented models of an industrial robotic swarm application and illustrate the use of AO modeling techniques to manage the complexity. We also show how to use AO modeling for design-space exploration. Read more

Download here

Control improvisation with probabilistic temporal specifications

Published in 2016 IEEE First International Conference on Internet-of-Things Design and Implementation (IoTDI), 2016

We consider the problem of generating randomized control sequences for complex networked systems typically actuated by human agents. Our approach leverages a concept known as control improvisation, which is based on a combination of data-driven learning and controller synthesis from formal specifications. We learn from existing data a generative model (for instance, an explicit-duration hidden Markov model, or EDHMM) and then supervise this model in order to guarantee that the generated sequences satisfy some desirable specifications given in Probabilistic Computation Tree Logic (PCTL). We present an implementation of our approach and apply it to the problem of mimicking the use of lighting appliances in a residential unit, with potential applications to home security and resource management. We present experimental results showing that our approach produces realistic control sequences, similar to recorded data based on human actuation, while satisfying suitable formal requirements. Read more

Download here

Coordinated actors for reliable self-adaptive systems

Published in International Workshop on Formal Aspects of Component Software, 2016

Self-adaptive systems are systems that automatically adapt in response to environmental and internal changes, such as possible failures and variations in resource availability. Such systems are often realized by a MAPE-K feedback loop, where Monitor, Analyze, Plan and Execute components have access to a runtime model of the system and environment which is kept in the Knowledge component. In order to provide guarantees on the correctness of a self-adaptive system at runtime, the MAPE-K feedback loop needs to be extended with assurance techniques. To address this issue, we propose a coordinated actor-based approach to build a reusable and scalable model@runtime for self-adaptive systems in the domain of track-based traffic control systems. We demonstrate the approach by implementing an automated Air Traffic Control system (ATC) using Ptolemy tool. We compare different adaptation policies on the ATC model based on performance metrics and analyze combination of policies in different configurations of the model. We enriched our framework with runtime performance analysis such that for any unexpected change, subsequent behavior of the model is predicted and results are used for adaptation at the change-point. Moreover, the developed framework enables checking safety properties at runtime. Read more

Download here

Information seeking and model predictive control of a cooperative multi-robot system

Published in Journal of Artificial Life and Robotics, 2016

In this paper, we propose a cooperative multi-robot control system, operating in an unfamiliar or unstructured environment. We focus on a robust model predictive control (robust-MPC) framework that enables robotic agents to operate in uncertain environments, and study the effect of observation uncertainties that arise from sensor noise on cooperative control performance. The proposed system relies on cooperative observation based on an information-seeking theory, in which the system not only can compensate uncertainty, but also takes actions to mitigate it. We carry out a case study that demonstrates a multi-robot collision avoidance scenario in an unknown environment. Simulation results show that the combination of robust-MPC methods and cooperative observation enables the cooperative multi-robot system to move efficiently and reach the goal faster than an uncooperative scenario. Read more

Download here

Cooperative multi-robot information acquisition based on distributed robust model predictive control

Published in 2016 IEEE International Conference on Robotics and Biomimetics (ROBIO), 2016

In this paper, we propose a distributed multi-robot control system working in dynamic and uncertain environments. Robust model predictive control (robust MPC) enables robots to deal with uncertainties. However, the performance of the robust MPC is dependent on the amount of uncertainty that derives from noisy measurements, communication disturbance, etc. The proposed system includes multiple observation robots that gather information cooperatively as well as a main robot controlled by robust MPC. Therefore, the system works for not only treating the uncertainty but also decreasing it. A simulation result of a collision avoidance shows that the information acquisition by the observation robots enables the main robot to move efficiently and arrive at the goal faster than a case without the observation robots. We also focus on a problem that a large number of observation robots will increase the frequency of inter-robot collision avoidances, and thus negatively affect to the performance of the main robot. Simulation results under various conditions on a disturbance level and a measurement range of sensors clarifies an adequate number of observation robots as well as the design guideline about sensors and networks. Read more

Download here

Specification mining for machine improvisation with formal specifications

Published in Computers in Entertainment (CIE), 2016

We address the problem of mining musical specifications from a training set of songs and using these specifications in a machine improvisation system capable of generating improvisations imitating a given style of music. Our inspiration comes from Control Improvisation, which combines learning and synthesis from formal specifications. We mine specifications from symbolic musical data with musical and general usage patterns. We use the mined specifications to ensure that an improvised musical sequence satisfies desirable properties given a harmonic context and phrase structure. We present a specification mining strategy based on pattern graphs and apply it to the problem of supervising the improvisation of blues songs. We present an analysis of the mined specifications and compare the results of improvisations generated with and without specifications. Read more

Download here

Coordinated actor model of self-adaptive track-based traffic control systems

Published in Journal of Systems and Software, 2018

Self-adaptation is a well-known technique to handle growing complexities of software systems, where a system autonomously adapts itself in response to changes in a dynamic and unpredictable environment. With the increasing need for developing self-adaptive systems, providing a model and an implementation platform to facilitate integration of adaptation mechanisms into the systems and assuring their safety and quality is crucial. In this paper, we target Track-based Traffic Control Systems (TTCSs) in which the traffic flows through pre-specified sub-tracks and is coordinated by a traffic controller. We introduce a coordinated actor model to design self-adaptive TTCSs and provide a general mapping between various TTCSs and the coordinated actor model. The coordinated actor model is extended to build large-scale self-adaptive TTCSs in a decentralized setting. We also discuss the benefits of using Ptolemy II as a framework for model-based development of large-scale self-adaptive systems that supports designing multiple hierarchical MAPE-K feedback loops interacting with each other. We propose a template based on the coordinated actor model to design a self-adaptive TTCS in Ptolemy II that can be instantiated for various TTCSs. We enhance the proposed template with a predictive adaptation feature. We illustrate applicability of the coordinated actor model and consequently the proposed template by designing two real-life case studies in the domains of air traffic control systems and railway traffic control systems in Ptolemy II. Read more

Download here

Solving Rubik’s Cube with a Robot Hand

Published in ArXiV, 2019

We demonstrate that models trained only in simulation can be used to solve a manipulation problem of unprecedented complexity on a real robot. This is made possible by two key components: a novel algorithm, which we call automatic domain randomization (ADR) and a robot platform built for machine learning. ADR automatically generates a distribution over randomized environments of ever-increasing difficulty. Control policies and vision state estimators trained with ADR exhibit vastly improved sim2real transfer. For control policies, memory-augmented models trained on an ADR-generated distribution of environments show clear signs of emergent meta-learning at test time. The combination of ADR with our custom robot platform allows us to solve a Rubik’s cube with a humanoid robot hand, which involves both control and state estimation problems. Read more

Download here

talks

teaching