arXiv subjects

Soumya Vasisht

Publications and source records attributed to Soumya Vasisht.

7 recordsLinked to original sources

arXiv2023

Mitigating Landside Congestion at Airports through Predictive Control of Diversionary Messages

We present a data-driven control framework for adaptively managing landside congestion at airports. Ground traffic significantly impacts airport operations and critical efficiency, environmental, and safety metrics. Our framework models a real-world traffic intervention currently deployed at Seattle-Tacoma International Airport (SEA), where a digital signboard recommends drivers to divert to Departures or Arrivals depending on current congestion. We use measured vehicle flow/speed and passenger volume data, as well as time-stamped records of diversionary messages, to build a macroscopic system dynamics model. We then design a model predictive controller that uses our estimated dynamics to recommend diversions that optimize for overall congestion. Finally, we evaluate our approach on 50 real-world historical scenarios at SEA where no diversions were deployed despite significant congestion. Our results suggest that our framework would have improved speed in the congested roadway by up to three times and saved between 20 and 80 vehicle-hours of cumulative travel time in every hour of deployment. Overall, our work emphasizes the opportunity of algorithmic decision-making to augment operator judgment and intuition and yield better real-world outcomes.

math.OC↗

arXiv2022

Estimating Driver Response Rates to Variable Message Signage at Seattle-Tacoma International Airport

We apply Bayesian Linear Regression to estimate the response rate of drivers to variable message signs at Seattle-Tacoma International Airport, or SeaTac. Our approach uses vehicle speed and flow data measured at the entrances of the arrival and departure-ways of the airport terminal, and sign message data. Depending on the time of day, we estimate that between 5.5 and 9.1% of drivers divert from departures to arrivals when the sign reads departures full, use arrivals, and conversely, between 1.9 and 4.2% of drivers divert from arrivals to departures. Though we lack counterfactual data (i.e., what would have happened had the diversionary treatment not been active), adopting a causal model that encodes time dependency with prior distributions rate can yield a measurable effect.

math.OC↗

arXiv2022

Domain-aware Control-oriented Neural Models for Autonomous Underwater Vehicles

Conventional physics-based modeling is a time-consuming bottleneck in control design for complex nonlinear systems like autonomous underwater vehicles (AUVs). In contrast, purely data-driven models, though convenient and quick to obtain, require a large number of observations and lack operational guarantees for safety-critical systems. Data-driven models leveraging available partially characterized dynamics have potential to provide reliable systems models in a typical data-limited scenario for high value complex systems, thereby avoiding months of expensive expert modeling time. In this work we explore this middle-ground between expert-modeled and pure data-driven modeling. We present control-oriented parametric models with varying levels of domain-awareness that exploit known system structure and prior physics knowledge to create constrained deep neural dynamical system models. We employ universal differential equations to construct data-driven blackbox and graybox representations of the AUV dynamics. In addition, we explore a hybrid formulation that explicitly models the residual error related to imperfect graybox models. We compare the prediction performance of the learned models for different distributions of initial conditions and control inputs to assess their accuracy, generalization, and suitability for control.

math.DS↗

arXiv2022

Dissipative Deep Neural Dynamical Systems

In this paper, we provide sufficient conditions for dissipativity and local asymptotic stability of discrete-time dynamical systems parametrized by deep neural networks. We leverage the representation of neural networks as pointwise affine maps, thus exposing their local linear operators and making them accessible to classical system analytic and design methods. This allows us to "crack open the black box" of the neural dynamical system's behavior by evaluating their dissipativity, and estimating their stationary points and state-space partitioning. We relate the norms of these local linear operators to the energy stored in the dissipative system with supply rates represented by their aggregate bias terms. Empirically, we analyze the variance in dynamical behavior and eigenvalue spectra of these local linear operators with varying weight factorizations, activation functions, bias terms, and depths.

cs.LG↗

arXiv2021

Deep Learning Explicit Differentiable Predictive Control Laws for Buildings

We present a differentiable predictive control (DPC) methodology for learning constrained control laws for unknown nonlinear systems. DPC poses an approximate solution to multiparametric programming problems emerging from explicit nonlinear model predictive control (MPC). Contrary to approximate MPC, DPC does not require supervision by an expert controller. Instead, a system dynamics model is learned from the observed system's dynamics, and the neural control law is optimized offline by leveraging the differentiable closed-loop system model. The combination of a differentiable closed-loop system and penalty methods for constraint handling of system outputs and inputs allows us to optimize the control law's parameters directly by backpropagating economic MPC loss through the learned system model. The control performance of the proposed DPC method is demonstrated in simulation using learned model of multi-zone building thermal dynamics.

eess.SY↗

arXiv2021

Constrained Block Nonlinear Neural Dynamical Models

Neural network modules conditioned by known priors can be effectively trained and combined to represent systems with nonlinear dynamics. This work explores a novel formulation for data-efficient learning of deep control-oriented nonlinear dynamical models by embedding local model structure and constraints. The proposed method consists of neural network blocks that represent input, state, and output dynamics with constraints placed on the network weights and system variables. For handling partially observable dynamical systems, we utilize a state observer neural network to estimate the states of the system's latent dynamics. We evaluate the performance of the proposed architecture and training methods on system identification tasks for three nonlinear systems: a continuous stirred tank reactor, a two tank interacting system, and an aerodynamics body. Models optimized with a few thousand system state observations accurately represent system dynamics in open loop simulation over thousands of time steps from a single set of initial conditions. Experimental results demonstrate an order of magnitude reduction in open-loop simulation mean squared error for our constrained, block-structured neural models when compared to traditional unstructured and unconstrained neural network models.

math.DS↗

arXiv2013

Mission to the Trojan Asteroids: lessons learned during a JPL Planetary Science Summer School mission design exercise

The 2013 Planetary Science Decadal Survey identified a detailed investigation of the Trojan asteroids occupying Jupiter's L4 and L5 Lagrange points as a priority for future NASA missions. Observing these asteroids and measuring their physical characteristics and composition would aid in identification of their source and provide answers about their likely impact history and evolution, thus yielding information about the makeup and dynamics of the early Solar System. We present a conceptual design for a mission to the Jovian Trojan asteroids: the Trojan ASteroid Tour, Exploration, and Rendezvous (TASTER) mission, that is consistent with the NASA New Frontiers candidate mission recommended by the Decadal Survey and the final result of the 2011 NASA-JPL Planetary Science Summer School. Our proposed mission includes visits to two Trojans in the L4 population: a 500 km altitude fly-by of 1999 XS143, followed by a rendezvous with and detailed observations of 911 Agamemnon at orbital altitudes of 1000 - 100 km over a 12 month nominal science data capture period. Our proposed instrument payload - wide- and narrow-angle cameras, a visual and infrared mapping spectrometer, and a neutron/gamma ray spectrometer - would provide unprecedented high-resolution, regional-to-global datasets for the target bodies, yielding fundamental information about the early history and evolution of the Solar System. Although our mission design was completed as part of an academic exercise, this study serves as a useful starting point for future Trojan mission design studies. In particular, we identify and discuss key issues that can make large differences in the complex trade-offs required when designing a mission to the Trojan asteroids.

astro-ph.EP↗