Matanya B. Horowitz

Linear Hamilton Jacobi Bellman Equations in high dimensions

The Hamilton Jacobi Bellman Equation (HJB) provides the globally optimal solution to large classes of control problems. Unfortunately, this generality comes at a price, the calculation of such solutions is typically intractible for systems with more than moderate state space size due to the curse of dimensionality. This work combines recent results in the structure of the HJB, and its reduction to a linear Partial Differential Equation (PDE), with methods based on low rank tensor representations, known as a separated representations, to address the curse of dimensionality. The result is an algorithm to solve optimal control problems which scales linearly with the number of states in a system, and is applicable to systems that are nonlinear with stochastic forcing in finite-horizon, average cost, and first-exit settings. The method is demonstrated on inverted pendulum, VTOL aircraft, and quadcopter models, with system dimension two, six, and twelve respectively.

Interactive non-prehensile manipulation for grasping via POMDPs

This paper develops a technique for an autonomous robot endowed with a manipulator arm, a multi-fingered gripper, and a variety of sensors to manipulate a known, but poorly observable object. We present a novel grasp planning method which not only incorporates the potential collision between object and manipulator, but takes advantage of this interaction. The natural uncertainty and difficulties in observation in such tasks is modeled as a Partially Observable Markov Decision Process (POMDP). Recent advances in point-based methods as well as a novel state space representation specific to the grasping problem are leveraged to overcome state space growth issues. Simulation results are presented for the combined localization, manipulation, and grasping of a small nut on a table.

Semidefinite Relaxations for Stochastic Optimal Control Policies

Recent results in the study of the Hamilton Jacobi Bellman (HJB) equation have led to the discovery of a formulation of the value function as a linear Partial Differential Equation (PDE) for stochastic nonlinear systems with a mild constraint on their disturbances. This has yielded promising directions for research in the planning and control of nonlinear systems. This work proposes a new method obtaining approximate solutions to these linear stochastic optimal control (SOC) problems. A candidate polynomial with variable coefficients is proposed as the solution to the SOC problem. A Sum of Squares (SOS) relaxation is then taken to the partial differential constraints, leading to a hierarchy of semidefinite relaxations with improving sub-optimality gap. The resulting approximate solutions are shown to be guaranteed over- and under-approximations for the optimal value function.

Convex relaxations of SE(2) and SE(3) for visual pose estimation

This paper proposes a new method for rigid body pose estimation based on spectrahedral representations of the tautological orbitopes of SE(2) and SE(3). The approach can use dense point cloud data from stereo vision or an RGB-D sensor (such as the Microsoft Kinect), as well as visual appearance data as input. The method is a convex relaxation of the classical pose estimation problem, and is based on explicit linear matrix inequality (LMI) representations for the convex hulls of SE(2) and SE(3). Given these representations, the relaxed pose estimation problem can be framed as a robust least squares problem with the optimization variable constrained to these convex sets. Although this formulation is a relaxation of the original problem, numerical experiments indicates that it is indeed exact - i.e. its solution is a member of SE(2) or SE(3) - in many interesting settings. We additionally show that this method is guaranteed to be exact for a large class of pose estimation problems.

Model-based autonomous system for performing dexterous, human-level manipulation tasks

This article presents a model based approach to autonomous dexterous manipulation, developed as part of the DARPA Autonomous Robotic Manipulation Software (ARM-S) program. Performing human-level manipulation tasks is achieved through a novel combination of perception in uncertain environments, precise tool use, forceful dual-arm planning and control, persistent environmental tracking, and task level verification. Deliberate interaction with the environment is incorporated into planning and control strategies, which, when coupled with world estimation, allows for refinement of models and precise manipulation. The system takes advantage of sensory feedback immediately with little open-loop execution, attempting true autonomous reasoning and multi-step sequencing that adapts in the face of changing and uncertain environments. A tire change scenario utilizing human tools, discussed throughout the article, is used to described the system approach. A second scenario of cutting a wire is also presented, and is used to illustrate system component reuse and generality.

A Compositional Approach to Stochastic Optimal Control with Co-safe Temporal Logic Specifications

We introduce an algorithm for the optimal control of stochastic nonlinear systems subject to temporal logic constraints on their behavior. We compute directly on the state space of the system, avoiding the expensive pre-computation of a discrete abstraction. An automaton that corresponds to the temporal logic specification guides the computation of a control policy that maximizes the probability that the system satisfies the specification. This reduces controller synthesis to solving a sequence of stochastic constrained reachability problems. Each individual reachability problem is solved via the Hamilton-Jacobi-Bellman (HJB) partial differential equation of stochastic optimal control theory. To increase the efficiency of our approach, we exploit a class of systems where the HJB equation is linear due to structural assumptions on the noise. The linearity of the partial differential equation allows us to pre-compute control policy primitives and then compose them, at essentially zero cost, to conservatively satisfy a complex temporal logic specification.

A convex approach to consensus on SO(n)

This paper introduces several new algorithms for consensus over the special orthogonal group. By relying on a convex relaxation of the space of rotation matrices, consensus over rotation elements is reduced to solving a convex problem with a unique global solution. The consensus protocol is then implemented as a distributed optimization using (i) dual decomposition, and (ii) both semi and fully distributed variants of the alternating direction method of multipliers technique - all with strong convergence guarantees. The convex relaxation is shown to be exact at all iterations of the dual decomposition based method, and exact once consensus is reached in the case of the alternating direction method of multipliers. Further, analytic and/or efficient solutions are provided for each iteration of these distributed computation schemes, allowing consensus to be reached without any online optimization. Examples in satellite attitude alignment with up to 100 agents, an estimation problem from computer vision, and a rotation averaging problem on SO(6) validate the approach.

Combined grasp and manipulation planning as a trajectory optimization problem

Many manipulation planning problems involve several related sub-problems, such as the selection of grasping points on an object, choice of hand posture, and determination of the arms configuration and evolving trajectory. Traditionally, these planning sub-problems have been handled separately, potentially leading to sub-optimal, or even infeasible, combinations of the individually determined solutions. This paper formulates the combined problem of grasp contact selection, grasp force optimization, and manipulator arm/hand trajectory planning as a problem in optimal control. That is, the locally optimal trajectory for the manipulator, hand mechanism, and contact locations are determined during the pre-grasping, grasping, and subsequent object transport phase. Additionally, a barrier function approach allows for non-feasible grasps to be optimized, enlarging the region of convergence for the algorithm. A simulation of a simple planar object manipulation task is used to illustrate and validate the approach.

Domain decomposition for stochastic optimal control

This work proposes a method for solving linear stochastic optimal control (SOC) problems using sum of squares and semidefinite programming. Previous work had used polynomial optimization to approximate the value function, requiring a high polynomial degree to capture local phenomena. To improve the scalability of the method to problems of interest, a domain decomposition scheme is presented. By using local approximations, lower degree polynomials become sufficient, and both local and global properties of the value function are captured. The domain of the problem is split into a non-overlapping partition, with added constraints ensuring C1 continuity. The Alternating Direction Method of Multipliers (ADMM) is used to optimize over each domain in parallel and ensure convergence on the boundaries of the partitions. This results in improved conditioning of the problem and allows for much larger and more complex problems to be addressed with improved performance.

Resilience in Large Scale Distributed Systems

Distributed systems are comprised of multiple subsystems that interact in two distinct ways: (1) physical interactions and (2) cyber interactions; i.e. sensors, actuators and computers controlling these subsystems, and the network over which they communicate. A broad class of cyber-physical systems (CPS) are described by such interactions, such as the smart grid, platoons of autonomous vehicles and the sensorimotor system. This paper will survey recent progress in developing a coherent mathematical framework that describes the rich CPS “design space” of fundamental limits and tradeoffs between efficiency, robustness, adaptation, verification and scalability. Whereas most research treats at most one of these issues, we attempt a holistic approach in examining these metrics. In particular, we will argue that a control architecture that emphasizes scalability leads to improvements in robustness, adaptation, and verification, all the while having only minor effects on efficiency – i.e. through the choice of a new architecture, we believe that we are able to bring a system closer to the true fundamental hard limits of this complex design space.