https://hdl.handle.net/1721.1/37334 Techincal reports developed by laboratory members Fri, 03 Apr 2026 18:40:02 GMT 2026-04-03T18:40:02Z https://hdl.handle.net/1721.1/88969 Learning Sparse Gaussian Graphical Model with l0-regularization Mu, Beipeng; How, Jonathan For the problem of learning sparse Gaussian graphical models, it is desirable to obtain both sparse structures as well as good parameter estimates. Classical techniques, such as optimizing the l1-regularized maximum likelihood or Chow-Liu algorithm, either focus on parameter estimation or constrain to speci c structure. This paper proposes an alternative that is based on l0-regularized maximum likelihood and employs a greedy algorithm to solve the optimization problem. We show that, when the graph is acyclic, the greedy solution finds the optimal acyclic graph. We also show it can update the parameters in constant time when connecting two sub-components, thus work efficiently on sparse graphs. Empirical results are provided to demonstrate this new algorithm can learn sparse structures with cycles efficiently and that it dominates l1-regularized approach on graph likelihood. Fri, 22 Aug 2014 00:00:00 GMT https://hdl.handle.net/1721.1/88969 2014-08-22T00:00:00Z https://hdl.handle.net/1721.1/77933 Supplementary material for nonparameteric adaptive control of time varying systems using gaussian processes Chowdhary, Girish; Kingravi, Hassan A.; How, Jonathan P.; Vela, Patricio A. Real-world dynamical variations make adaptive control of time-varying systems highly relevant. However, most adaptive control literature focuses on analyzing systems where the uncertainty is represented as a weighted linear combination of fixed number of basis functions, with constant weights. One approach to modeling time variations is to assume time varying ideal weights, and use difference integration to accommodate weight variation. However, this approach reactively suppresses the uncertainty, and has little ability to predict system behavior locally. We present an alternate formulation by leveraging Bayesian nonparametric Gaussian Process adaptive elements. We show that almost surely bounded adaptive controllers for a class of nonlinear time varying system can be formulated by incorporating time as an additional input to the Gaussian kernel. Analysis and simulations show that the learning-enabled local predictive ability of our adaptive controllers significantly improves performance. This technical report has supplementary material for "Bayesian Nonparametric Adaptive Control of Time-varying Systems using Gaussian Processes" American Control Conference paper Fri, 15 Mar 2013 00:00:00 GMT https://hdl.handle.net/1721.1/77933 2013-03-15T00:00:00Z https://hdl.handle.net/1721.1/77931 Bayesian Nonparametric Adaptive Control using Gaussian Processes Chowdhary, Girish; Kingravi, Hassan A.; How, Jonathan P.; Vela, Patricio A. Most current Model Reference Adaptive Control (MRAC) methods rely on parametric adaptive elements, in which the number of parameters of the adaptive element are fixed a priori, often through expert judgment. An example of such an adaptive element are Radial Basis Function Networks (RBFNs), with RBF centers pre-allocated based on the expected operating domain. If the system operates outside of the expected operating domain, this adaptive element can become non-effective in capturing and canceling the uncertainty, thus rendering the adaptive controller only semi-global in nature. This paper investigates a Gaussian Process (GP) based Bayesian MRAC architecture (GP-MRAC), which leverages the power and flexibility of GP Bayesian nonparametric models of uncertainty. GP-MRAC does not require the centers to be preallocated, can inherently handle measurement noise, and enables MRAC to handle a broader set of uncertainties, including those that are defined as distributions over functions. We use stochastic stability arguments to show that GP-MRAC guarantees good closed loop performance with no prior domain knowledge of the uncertainty. Online implementable GP inference methods are compared in numerical simulations against RBFN-MRAC with preallocated centers and are shown to provide better tracking and improved long-term learning. This technical report is a preprint of an article submitted to a journal. Fri, 15 Mar 2013 00:00:00 GMT https://hdl.handle.net/1721.1/77931 2013-03-15T00:00:00Z https://hdl.handle.net/1721.1/77915 Efficient Distributed Sensing Using Adaptive Censoring-Based Inference Mu, Beipeng; Chowdhary, Girish; How, Jonathan P. In many distributed sensing applications it is likely that only a few agents will have valuable information at any given time. Since wireless communication between agents is resource-intensive, it is important to ensure that the communication effort is focused on communicating valuable information from informative agents. This paper presents communication efficient distributed sensing algorithms that avoid network cluttering by having only agents with high Value of Information (VoI) broadcast their measurements to the network, while others censor themselves. A novel contribution of the presented distributed estimation algorithm is the use of an adaptively adjusted VoI threshold to determine which agents are informative. This adaptation enables the team to better balance between the communication cost incurred and the long-term accuracy of the estimation. Theoretical results are presented establishing the almost sure convergence of the communication cost and estimation error to zero for distributions in the exponential family. Furthermore, validation through numerical simulations and real datasets show that the new VoI-based algorithms can yield improved parameter estimates than those achieved by previously published hyperparameter consensus algorithms while incurring only a fraction of the communication cost. This technical report is a preprint of work submitted to a journal. Fri, 15 Mar 2013 00:00:00 GMT https://hdl.handle.net/1721.1/77915 2013-03-15T00:00:00Z https://hdl.handle.net/1721.1/71875 Efficient distributed information fusion using value of information based censoring Mu, Beipeng; How, Jonathan P.; Chowdhary, Girish In many distributed sensing applications, not all agents have valuable information at all times. Therefore, requiring all agents to communicate at all times can be resource intensive. In this work, the notion of Value of Information (VoI) is used to improve the efficiency of distributed sensing algorithms. Particularly, only agents with high VoI broadcast their measurements to the network, while others censor their measurements. New VoI realized data fusion algorithms are introduced, and an in depth analysis of the costs incurred by these algorithms and conventional distributed data fusion algorithms is presented. Numerical simulations are used to compare the performance of the VoI realized algorithms with traditional data fusion algorithms. A VoI based algorithm that adaptively adjusts the criterion for being informative is presented and shown to strike a good balance between reduced communication cost and increased accuracy. Fri, 27 Jul 2012 00:00:00 GMT https://hdl.handle.net/1721.1/71875 2012-07-27T00:00:00Z https://hdl.handle.net/1721.1/64774 Threat Assessment Design for Driver Assistance System at Intersections: Experiment Video Aoude, Georges S.; Luders, Brandon D.; Lee, Kenneth K. H.; Levine, Daniel S.; How, Jonathan P. Thu, 07 Jul 2011 00:00:00 GMT https://hdl.handle.net/1721.1/64774 2011-07-07T00:00:00Z https://hdl.handle.net/1721.1/64738 Probabilistically Safe Avoidance of Dynamic Obstacles with Uncertain Motion Patterns Luders, Brandon D.; Aoude, Georges S.; Joseph, Joshua M.; Roy, Nicholas; How, Jonathan P. This paper presents a real-time path planning algorithm which can guarantee probabilistic feasibility for autonomous robots subject to process noise and an uncertain environment, including dynamic obstacles with uncertain motion patterns. The key contribution of the work is the integration of a novel method for modeling dynamic obstacles with uncertain future trajectories. The method, denoted as RR-GP, uses a learned motion pattern model of the dynamic obstacles to make long-term predictions of their future paths. This is done by combining the flexibility of Gaussian processes (GP) with the efficiency of RRT-Reach, a sampling-based reachability computation method which ensures dynamic feasibility. This prediction model is then utilized within chance-constrained rapidly-exploring random trees (CC-RRT), which uses chance constraints to explicitly achieve probabilistic constraint satisfaction while maintaining the computational benefits of sampling-based algorithms. With RR-GP embedded in the CC-RRT framework, theoretical guarantees can be demonstrated for linear systems subject to Gaussian uncertainty, though the extension to nonlinear systems is also considered. Simulation results show that the resulting approach can be used in real-time to efficiently and accurately execute safe paths. Mon, 04 Jul 2011 00:00:00 GMT https://hdl.handle.net/1721.1/64738 2011-07-04T00:00:00Z https://hdl.handle.net/1721.1/64737 Threat-aware Path Planning in Uncertain Urban Environments [Attached Video] Aoude, Georges S.; Luders, Brandon D.; Levine, Daniel S.; How, Jonathan P. This paper considers the path planning problem for an autonomous vehicle in an urban environment populated with static obstacles and moving vehicles with uncertain intents. We propose a novel threat assessment module, consisting of an intention predictor and a threat assessor, which augments the host vehicle's path planner with a real-time threat value representing the risks posed by the estimated intentions of other vehicles. This new threat-aware planning approach is applied to the CL-RRT path planning framework, used by the MIT team in the 2007 DARPA Grand Challenge. The strengths of this approach are demonstrated through simulation and experiments performed in the RAVEN testbed facilities. Fri, 01 Oct 2010 00:00:00 GMT https://hdl.handle.net/1721.1/64737 2010-10-01T00:00:00Z https://hdl.handle.net/1721.1/56001 Corrections to "Geometric Properties of Gradient Projection Anti-windup Compensated Systems" Teo, Justin; How, Jonathan P. In a conference paper titled "Geometric Properties of Gradient Projection Anti-windup Compensated Systems," two main results were presented. The first is the controller state-output consistency property of gradient projection anti-windup (GPAW) compensated controllers. The second is a geometric bounding condition relating the vector fields of the uncompensated and GPAW compensated closed-loop systems with respect to a star domain. While the controller state-output consistency property stands without modifications, the proof of the geometric bounding condition depends on two lemmas, the proofs of which were found to be faulty. In this report, we present a new proof of the geometric bounding condition using concepts from convex analysis, together with minor miscellaneous corrections. Tue, 29 Jun 2010 00:00:00 GMT https://hdl.handle.net/1721.1/56001 2010-06-29T00:00:00Z https://hdl.handle.net/1721.1/52600 Gradient Projection Anti-windup Scheme on Constrained Planar LTI Systems Teo, Justin; How, Jonathan P. The gradient projection anti-windup (GPAW) scheme was recently proposed as an anti-windup method for nonlinear multi-input-multi-output systems/controllers, the solution of which was recognized as a largely open problem in a recent survey paper. This report analyzes the properties of the GPAW scheme applied to an input constrained first order linear time invariant (LTI) system driven by a first order LTI controller, where the objective is to regulate the system state about the origin. We show that the GPAW compensated system is in fact a projected dynamical system (PDS), and use results in the PDS literature to assert existence and uniqueness of its solutions. The main result is that the GPAW scheme can only maintain/enlarge the exact region of attraction of the uncompensated system. We illustrate the qualitative weaknesses of some results in establishing true advantages of anti-windup methods, and propose a new paradigm to address the anti-windup problem, where results relative to the uncompensated system are sought. Mon, 15 Mar 2010 20:16:57 GMT https://hdl.handle.net/1721.1/52600 2010-03-15T20:16:57Z https://hdl.handle.net/1721.1/46720 Using Support Vector Machines and Bayesian Filtering for Classifying Agent Intentions at Road Intersections Aoude, Georges S.; How, Jonathan P. Classifying other agents’ intentions is a very complex task but it can be very essential in assisting (autonomous or human) agents in navigating safely in dynamic and possibly hostile environments. This paper introduces a classification approach based on support vector machines and Bayesian filtering (SVM-BF). It then applies it to a road intersection problem to assist a vehicle in detecting the intention of an approaching suspicious vehicle. The SVM-BF approach achieved very promising results. Tue, 15 Sep 2009 22:21:04 GMT https://hdl.handle.net/1721.1/46720 2009-09-15T22:21:04Z https://hdl.handle.net/1721.1/45514 On Approximate Dynamic Inversion Teo, Justin; How, Jonathan P. Approximate Dynamic Inversion has been established as a method to control minimum-phase, nonaffine-in-control systems [1]. In this report, we re-state the main results of [1], clarify some minor notational errors, and prove the same results in an expanded form. In the large, the main results of [1] still stand. The development follows [1] closely, and no novelty is claimed herein. The purpose of this report is mainly to supplement our existing results in [2]–[4] that rely heavily on the results of [1]. Fri, 08 May 2009 15:45:31 GMT https://hdl.handle.net/1721.1/45514 2009-05-08T15:45:31Z https://hdl.handle.net/1721.1/43949 Reaching Consensus with Imprecise Probabilities over a Network Bertuccelli, Luca F.; How, Jonathan P. This paper discusses the problem of a distributed network of agents attempting to agree on an imprecise probability over a network. Unique from other related work however, the agents must reach agreement while accounting for relative uncertainties in their respective probabilities. First, we assume that the agents only seek to agree to a centralized estimate of the probabilities, without accounting for observed transitions. We provide two methods by which such an agreement can occur which uses ideas from Dirichlet distributions. The first methods interprets the consensus problem as an aggregation of Dirichlet distributions of the neighboring agents. The second method uses ideas from Kalman Consensus to approximate this consensus using the mean and the variance of the Dirichlet distributions. A key results of this paper is that we show that when the agents are simultaneously actively observing state transitions and attempting to reach consensus on the probabilities, the agreement protocol can be insensitive to any new information, and agreement is not possible. Ideas from exponential fading are adopted to improve convergence and reach a consistent agreement. Sat, 20 Dec 2008 12:46:21 GMT https://hdl.handle.net/1721.1/43949 2008-12-20T12:46:21Z https://hdl.handle.net/1721.1/42839 Equivalence between Approximate Dynamic Inversion and Proportion-Integral Control Teo, Justin; How, Jonathan P. Approximate Dynamic Inversion (ADI) has been established as a method to control minimum-phase, nonaffine-in-control systems. Previous results have shown that for single-input nonaffine-in-control systems, every ADI controller admits a linear Proportional-Integral (PI) realization that is largely independent of the nonlinear function that defines the system. In this report, we first present an extension of the ADI method for single-input nonaffine-in-control systems that renders the closed-loop error dynamics independent of the reference model dynamics. The equivalent PI controller will be derived and both of these results are then extended to multi-input nonaffine-in-control systems. Mon, 29 Sep 2008 14:05:49 GMT https://hdl.handle.net/1721.1/42839 2008-09-29T14:05:49Z https://hdl.handle.net/1721.1/37335 Hover, Transition, and Level Flight Control Design for a Single-Propeller Indoor Airplane Frank, Adrian; McGrew, James; Valenti, Mario; Levine, Daniel; How, Jonathan P. This paper presents vehicle models and test flight results for an autonomous fixed-wing airplane that is designed to take-off, hover, transition to and from level-flight modes, and perch on a vertical landing platform in a highly space constrained environment. By enabling a fixed-wing UAV to achieve these feats, the speed and range of a fixed-wing aircraft in level flight are complimented by hover capabilities that were typically limited to rotorcraft. Flight and perch landing results are presented. This capability significantly eases support and maintenance of the vehicle. All of the flights presented in this paper are performed using the MIT Real-time Autonomous Vehicle indoor test ENvironment (RAVEN). Tue, 15 May 2007 07:08:23 GMT https://hdl.handle.net/1721.1/37335 2007-05-15T07:08:23Z