Augustinos (Augustine) Saravanos

I’m a third year PhD in Machine Learning student at Georgia Tech. I am fortunate to be part of the Autonomous Control and Decision Systems Lab and to be advised by Prof. Evangelos Theodorou. I also work closely with Prof. Efstathios Bakolas from UT Austin.

My research focuses on developing distributed methods for the control of large-scale multi-agent systems . In particular, combining concepts from the areas of distributed optimization, stochastic optimal control and machine learning, my work aims in developing scalable and computationally efficient decentralized algorithms for controlling multi-agent systems with thousands of robots!

Some indicative works are Distributed Differential Dynamic Programming [Paper (preprint), Video], Decentralized Multi-Agent Deep FBSDEs [Paper (RSS ‘22), Video] and Distributed Covariance Steering [Paper (RSS ‘21) , Video].

Prior to Georgia Tech, I graduated (top 1%) with a Diploma in Electrical and Computer Engineering from the University of Patras in Greece, and was advised by Prof. Evangelos Papadopoulos and Prof. Nick Koussoulas.

See my full CV here.

Contact: asaravanos3 [at] gatech [dot] edu

Publications

JMLR
(In Submission)

Sampling-Based Optimization for Multi-Agent Model Predictive Control

Ziyi Wang, Augustinos D. Saravanos , Hassan Almubarak, Oswin So, and Evangelos Theodorou

Journal of Machine Learning Research (In Submission) , 2022

Abs PDF Video

We systematically review the Variational Optimization, Variational Inference and Stochastic Search perspectives on sampling-based dynamic optimization and discuss their connections to state-of-the-art optimizers and Stochastic Optimal Control (SOC) theory. A general convergence and sample complexity analysis on the three perspectives is provided through the unifying Stochastic Search perspective. We then extend these frameworks to their distributed versions for multi-agent control by combining them with consensus Alternating Direction Method of Multipliers (ADMM) to decouple the full problem into local neighborhood-level ones that can be solved in parallel. Model Predictive Control (MPC) algorithms are then developed based on these frameworks, leading to fully decentralized sampling-based dynamic optimizers. The capabilities of the proposed algorithms framework are demonstrated on multiple complex multi-agent tasks for vehicle and quadcopter systems in simulation. The results compare different distributed sampling-based optimizers and their centralized counterparts. \augustinosMaybe add different policies. The scalability of the proposed distributed algorithms is demonstrated on a 196-vehicle scenario where a direct application of centralized sampling-based methods is shown to be prohibitive.
ICRA 2023
(In Submission)

Distributed Model Predictive Covariance Steering

Augustinos D. Saravanos , Isin M Balci, Efstathios Bakolas, and Evangelos Theodorou

IEEE International Conference on Robotics and Automation (ICRA) 2023 (In Submission) , 2022

Abs PDF Video

This paper proposes Distributed Model Predictive Covariance Steering (DMPCS), a novel method for safe multi-robot control under uncertainty. The scope of our approach is to blend covariance steering theory, distributed optimization and model predictive control (MPC) into a single methodology that is safe, scalable and decentralized. Initially, we pose a problem formulation that uses the Wasserstein distance to steer the state distributions of a multi-robot team to desired targets, and probabilistic constraints to ensure safety. We then transform this problem into a finite-dimensional optimization one by utilizing a disturbance feedback policy parametrization for covariance steering and a tractable approximation of the safety constraints. To solve the latter problem, we derive a decentralized consensus-based algorithm using the Alternating Direction Method of Multipliers (ADMM). This method is then extended to a receding horizon form, which yields the proposed DMPCS algorithm. Simulation experiments on large-scale problems with up to hundreds of robots successfully demonstrate the effectiveness and scalability of DMPCS. Its superior capability in achieving safety is also highlighted through a comparison against a standard stochastic MPC approach.
IEEE T-RO
(In Submission)

Distributed Differential Dynamic Programming Architectures for Large-Scale Multi-Agent Control

Augustinos D. Saravanos , Yuichiro Aoyama, Hongchang Zhu, and Evangelos A Theodorou

IEEE Transactions on Robotics (In Submission) , 2022

Abs PDF Video

In this paper, we propose two novel decentralized optimization frameworks for multi-agent nonlinear optimal control problems in robotics. The aim of this work is to suggest architectures that inherit the computational efficiency and scalability of Differential Dynamic Programming (DDP) and the distributed nature of the Alternating Direction Method of Multipliers (ADMM). In this direction, two frameworks are introduced. The first one called Nested Distributed DDP (ND-DDP), is a three-level architecture which employs ADMM for enforcing a consensus between all agents, an Augmented Lagrangian layer for satisfying local constraints and DDP as each agent’s optimizer. In the second approach, both consensus and local constraints are handled with ADMM, yielding a two-level architecture called Merged Distributed DDP (MD-DDP), which further reduces computational complexity. Both frameworks are fully decentralized since all computations are parallelizable among the agents and only local communication is necessary. Simulation results that scale up to thousands of vehicles and hundreds of drones verify the effectiveness of the methods. Superior scalability to large-scale systems against centralized DDP and centralized/decentralized sequential quadratic programming is also illustrated. Finally, hardware experiments on a multi-robot platform demonstrate the applicability of the proposed algorithms, while highlighting the importance of optimizing for feedback policies to increase robustness against uncertainty. A video with all results is available here.
RSS 2022

Decentralized Safe Multi-agent Stochastic Optimal Control using Deep FBSDEs and ADMM

Marcus A. Pereira*, Augustinos D. Saravanos* , Oswin So, and Evangelos A. Theodorou

Robotics: Science and Systems 2022 (*Equal Contribution) , 2022

Abs PDF Video

In this work, we propose a novel safe and scalable decentralized solution for multi-agent control in the presence of stochastic disturbances. Safety is mathematically encoded using stochastic control barrier functions and safe controls are computed by solving quadratic programs. Decentralization is achieved by augmenting to each agent’s optimization variables, copy variables, for its neighboring agents. This allows us to decouple the centralized multi-agent optimization problem. However, to ensure safety, neighboring agents must agree on what is safe for both of us and this creates a need for consensus. To enable safe consensus solutions, we incorporate an ADMM-based approach. Specifically, we propose a Merged CADMM-OSQP implicit neural network layer, that solves a mini-batch of both, local quadratic programs as well as the overall consensus problem, as a single optimization problem. This layer is embedded within a Deep FBSDEs network architecture at every time step, to facilitate end-to-end differentiable, safe and decentralized stochastic optimal control. The efficacy of the proposed approach is demonstrated on several challenging multi-robot tasks in simulation. By imposing requirements on safety specified by collision avoidance constraints, the safe operation of all agents is ensured during the entire training process. We also demonstrate superior scalability in terms of computational and memory savings as compared to a centralized approach.
RSS 2022
SRL Workshop

Sim2Real on the Robotarium Platform Using Decentralized Multi-Agent Safe Deep FBSDEs

Marcus A. Pereira*, Augustinos D. Saravanos* , and Evangelos A. Theodorou

Robotics: Science and Systems 2022: Workshop on Scaling Robot Learning (*Equal Contribution) , 2022

Abs PDF Video Video 2

In this work, we propose a novel sim2real framework for multi-robot control that relies on stochastic optimal control theory. Specifically, we use a recently proposed Deep Forward-Backward Stochastic Differential Equations (FBSDEs) algorithm to train LSTM-network-based feedback policies in simulation that are subsequently deployed directly on real hardware. This particular Deep FBSDE variant is tailored for multi-agent systems and ensures safety during the entire training process. Safety is facilitated by employing stochastic control barrier functions and decentralization is achieved by an ADMM-based consensus optimization approach. By randomizing the initial conditions and inducing noise into the robot’s dynamics, the policy is trained to control a mini-batch of trajectories to their respective targets on average, while ensuring safety for each trajectory with probability 1. We hypothesize that the ability to control an ensemble of trajectories empowers the feedback policy to compensate for uncertainty when transferred from simulation to real robots. We test this hypothesis on the Robotarium swarm-robotics testbed and successfully demonstrate the completion of tasks as well as safe operation when a policy trained in simulation is deployed without any prior real-world experience.
RSS 2021

Distributed Covariance Steering with Consensus ADMM for Stochastic Multi-Agent Systems

Augustinos D. Saravanos , Alexandros Tsolovikos, Efstathios Bakolas, and Evangelos Theodorou

Proceedings of Robotics: Science and Systems , 2021

Abs PDF Video

In this paper, we address the problem of steering a team of agents under stochastic linear dynamics to prescribed final state means and covariances. The agents operate in a common environment where inter-agent constraints may also be present. In order for our method to be scalable to large-scale systems and computationally efficient, we approach the problem in a distributed control framework using the Alternating Direction Method of Multipliers (ADMM). Each agent solves its own covariance steering problem in parallel, while additional copy variables for its closest neighbors are introduced to ensure that the inter-agent constraints will be satisfied. The inclusion of these additional variables creates a requirement for consensus between original and copy variables that involve the same agent. For this reason, we employ a variation of ADMM for consensus optimization. Simulation results on multi-vehicle systems under uncertainty with collision avoidance constraints illustrate the effectiveness of our algorithm. The substantially improved scalability of our distributed approach with respect to the number of agents is also demonstrated, in comparison with an equivalent centralized scheme.
CDC 2021

Receding Horizon Differential Dynamic Programming Under Parametric Uncertainty

Yuichiro Aoyama, Augustinos D. Saravanos , and Evangelos A. Theodorou

2021 60th IEEE Conference on Decision and Control (CDC) , 2021

Abs PDF

Generalized Polynomial Chaos (gPC) theory has been widely used for representing parametric uncertainty in a system, thanks to its ability to propagate uncertainty evolution. In an optimal control context, gPC can be combined with several optimization techniques to achieve a control policy that handles effectively this type of uncertainty. Such a suitable method is Differential Dynamic Programming (DDP), leading to an algorithm that inherits the scalability to high-dimensional systems and fast convergence nature of the latter. In this paper, we expand this combination aiming to acquire probabilistic guarantees on the satisfaction of nonlinear constraints. In particular, we exploit the ability of gPC to express higher order moments of the uncertainty distribution - without any Gaussianity assumption - and we incorporate chance constraints that lead to expressions involving the state covariance. Furthermore, we demonstrate that by implementing our algorithm in a receding horizon fashion, we are able to compute control policies that effectively reduce the accumulation of uncertainty on the trajectory. The applicability of our method is verified through simulation results on a differential wheeled robot and a quadrotor that perform obstacle avoidance tasks.
University
of Patras

Nonlinear Model Predictive Control for Space Robotic Systems

Augustinos D. Saravanos

University of Patras , 2019

Abs PDF

The number of satellites orbiting around the earth is continuously increasing. All these satellites demand servicing and refueling, otherwise they can become space debris. Hence, the development of space manipulators that would capture, service and decommission these satellites has become urgent. This kind of space robotic systems present many complexities while their control design has to be robust, accurate and as computationally light as possible. In this context, the objective of this thesis is the development of a Model Predictive Controller (MPC) for a robotic spacecraft/manipulator robotic system. In particular, we aim to investigate the performance of established linear MPC methods on a complex nonlinear system such as the space robotic system of this thesis. A MPC requires a model of the system in order to generate predictions about the future behaviour of the system and decide the control input that will be applied. An efficient state estimator is essential in order for the MPCs that have been developed to function properly, thus two have been designed: an Extended Kalman Filter (EKF) and a Nonlinear Moving Horizon Estimation (NMHE) algorithm. The NMHE employs an optimization-based approach taking into account past measurements for a certain horizon and using a model based on which it calculates the state estimates that agree the most with the measurements. The MHE and MPC are closely related methods since they operate in a receding horizon fashion and require a model of the system. The first MPC algorithm that has been implemented is a basic Generalized Predictive Controller (GPC). The second one is more advanced and is referred to as Dual-Mode MPC. The first part of the simulation results illustrate that in most cases the NMHE algorithm is superior to the EKF. Therefore, the NMHE has been integrated to the MPCs that have been developed. The second part of the comparative study demonstrates a comparison between the Dual-Mode MPC, the GPC and a baseline MIMO PID controller that has been designed in order to be a basis of comparison with the two MPCs. The simulation results indicate that the Dual-Mode MPC is clearly more preferable than the other two controllers. Finally, we make some conclusions with the most important being that the Dual-Mode MPC was an effective controller for the complex space robotic system it was applied on, and therefore, we suggest the application of established linear MPC techniques on nonlinear systems.