arXiv subjects

Pan Zhao

Publications and source records attributed to Pan Zhao.

36 records · Page 2Linked to original sources

arXiv2023

Positivity-free Policy Learning with Observational Data

Policy learning utilizing observational data is pivotal across various domains, with the objective of learning the optimal treatment assignment policy while adhering to specific constraints such as fairness, budget, and simplicity. This study introduces a novel positivity-free (stochastic) policy learning framework designed to address the challenges posed by the impracticality of the positivity assumption in real-world scenarios. This framework leverages incremental propensity score policies to adjust propensity score values instead of assigning fixed values to treatments. We characterize these incremental propensity score policies and establish identification conditions, employing semiparametric efficiency theory to propose efficient estimators capable of achieving rapid convergence rates, even when integrated with advanced machine learning algorithms. This paper provides a thorough exploration of the theoretical guarantees associated with policy learning and validates the proposed framework's finite-sample performance through comprehensive numerical experiments, ensuring the identification of causal effects from observational data is both robust and reliable.

stat.ME↗

arXiv2023

Safe and Efficient Reinforcement Learning Using Disturbance-Observer-Based Control Barrier Functions

Safe reinforcement learning (RL) with assured satisfaction of hard state constraints during training has recently received a lot of attention. Safety filters, e.g., based on control barrier functions (CBFs), provide a promising way for safe RL via modifying the unsafe actions of an RL agent on the fly. Existing safety filter-based approaches typically involve learning of uncertain dynamics and quantifying the learned model error, which leads to conservative filters before a large amount of data is collected to learn a good model, thereby preventing efficient exploration. This paper presents a method for safe and efficient RL using disturbance observers (DOBs) and control barrier functions (CBFs). Unlike most existing safe RL methods that deal with hard state constraints, our method does not involve model learning, and leverages DOBs to accurately estimate the pointwise value of the uncertainty, which is then incorporated into a robust CBF condition to generate safe actions. The DOB-based CBF can be used as a safety filter with model-free RL algorithms by minimally modifying the actions of an RL agent whenever necessary to ensure safety throughout the learning process. Simulation results on a unicycle and a 2D quadrotor demonstrate that the proposed method outperforms a state-of-the-art safe RL algorithm using CBFs and Gaussian processes-based model learning, in terms of safety violation rate, and sample and computational efficiency.

cs.RO↗

arXiv2023

Tube-Certified Trajectory Tracking for Nonlinear Systems With Robust Control Contraction Metrics

This paper presents an approach towards guaranteed trajectory tracking for nonlinear control-affine systems subject to external disturbances based on robust control contraction metrics (CCM) that aims to minimize the $\mathcal L_\infty$ gain from the disturbances to nominal-actual trajectory deviations. The guarantee is in the form of invariant tubes, computed offline and valid for any nominal trajectories, in which the actual states and inputs of the system are guaranteed to stay despite disturbances. Under mild assumptions, we prove that the proposed robust CCM (RCCM) approach yields tighter tubes than an existing approach based on CCM and input-to-state stability analysis. We show how the RCCM-based tracking controller together with tubes can be incorporated into a feedback motion planning framework to plan safe trajectories for robotic systems. Simulation results illustrate the effectiveness of the proposed method and empirically demonstrate reduced conservatism compared to the CCM-based approach.

eess.SY↗

arXiv2023

Safety Embedded Stochastic Optimal Control of Networked Multi-Agent Systems via Barrier States

This paper presents a novel approach for achieving safe stochastic optimal control in networked multi-agent systems (MASs). The proposed method incorporates barrier states (BaSs) into the system dynamics to embed safety constraints. To accomplish this, the networked MAS is factorized into multiple subsystems, and each one is augmented with BaSs for the central agent. The optimal control law is obtained by solving the joint Hamilton-Jacobi-Bellman (HJB) equation on the augmented subsystem, which guarantees safety via the boundedness of the BaSs. The BaS-based optimal control technique yields safe control actions while maintaining optimality. The safe optimal control solution is approximated using path integrals. To validate the effectiveness of the proposed approach, numerical simulations are conducted on a cooperative UAV team in two different scenarios.

eess.SY↗

arXiv2023

Verification of $L_1$ Adaptive Control using Verse Library: A Case Study of Quadrotors

$L_1$ adaptive control ($L_1$AC) is a control design technique that can handle a broad class of system uncertainties and provide transient performance guarantees. In this work-in-progress abstract, we discuss how existing formal verification tools can be applied to check the performance of $L_1$AC systems. We show that the theoretical transient performance and robustness guarantees of an $L_1$ adaptive controller for an 18-dimensional quadrotor system can be verified using the recently developed Verse reachability analysis tool. We will further consider the performance verification of $L_1$AC on systems with learning-enabled components.

eess.SY↗

arXiv2023

Optimizing Crop Management with Reinforcement Learning and Imitation Learning

Crop management, including nitrogen (N) fertilization and irrigation management, has a significant impact on the crop yield, economic profit, and the environment. Although management guidelines exist, it is challenging to find the optimal management practices given a specific planting environment and a crop. Previous work used reinforcement learning (RL) and crop simulators to solve the problem, but the trained policies either have limited performance or are not deployable in the real world. In this paper, we present an intelligent crop management system which optimizes the N fertilization and irrigation simultaneously via RL, imitation learning (IL), and crop simulations using the Decision Support System for Agrotechnology Transfer (DSSAT). We first use deep RL, in particular, deep Q-network, to train management policies that require all state information from the simulator as observations (denoted as full observation). We then invoke IL to train management policies that only need a limited amount of state information that can be readily obtained in the real world (denoted as partial observation) by mimicking the actions of the previously RL-trained policies under full observation. We conduct experiments on a case study using maize in Florida and compare trained policies with a maize management guideline in simulations. Our trained policies under both full and partial observations achieve better outcomes, resulting in a higher profit or a similar profit with a smaller environmental impact. Moreover, the partial-observation management policies are directly deployable in the real world as they use readily available information.

cs.AI↗

arXiv2023

Efficient and robust transfer learning of optimal individualized treatment regimes with right-censored survival data

An individualized treatment regime (ITR) is a decision rule that assigns treatments based on patients' characteristics. The value function of an ITR is the expected outcome in a counterfactual world had this ITR been implemented. Recently, there has been increasing interest in combining heterogeneous data sources, such as leveraging the complementary features of randomized controlled trial (RCT) data and a large observational study (OS). Usually, a covariate shift exists between the source and target population, rendering the source-optimal ITR unnecessarily optimal for the target population. We present an efficient and robust transfer learning framework for estimating the optimal ITR with right-censored survival data that generalizes well to the target population. The value function accommodates a broad class of functionals of survival distributions, including survival probabilities and restrictive mean survival times (RMSTs). We propose a doubly robust estimator of the value function, and the optimal ITR is learned by maximizing the value function within a pre-specified class of ITRs. We establish the $N^{-1/3}$ rate of convergence for the estimated parameter indexing the optimal ITR, and show that the proposed optimal value estimator is consistent and asymptotically normal even with flexible machine learning methods for nuisance parameter estimation. We evaluate the empirical performance of the proposed method by simulation studies and a real data application of sodium bicarbonate therapy for patients with severe metabolic acidaemia in the intensive care unit (ICU), combining a RCT and an observational study with heterogeneity.

stat.ME↗

arXiv2022

$\mathcal{L}_1$ Adaptive Control with Switched Reference Models: Application to Learn-to-Fly

Learn-to-Fly (L2F) is a new framework that aims to replace the traditional iterative development paradigm for aerial vehicles with a combination of real-time aerodynamic modeling, guidance, and learning control. To ensure safe learning of the vehicle dynamics on the fly, this paper presents an $\mathcal{L}_1$ adaptive control ($\mathcal{L}_1$AC) based scheme, which actively estimates and compensates for the discrepancy between the intermediately learned dynamics and the actual dynamics. First, to incorporate the periodic update of the learned model within the L2F framework, this paper extends the $\mathcal{L}_1$AC architecture to handle a switched reference system subject to unknown time-varying parameters and disturbances. The paper also includes an analysis of both transient and steady-state performance of the $\mathcal{L}_1$AC architecture in the presence of non-zero initialization error for the state predictor. Second, the paper presents how the proposed $\mathcal{L}_1$AC scheme is integrated into the L2F framework, including its interaction with the baseline controller and the real-time modeling module. Finally, flight tests on an unmanned aerial vehicle (UAV) validate the efficacy of the proposed control and learning scheme.

eess.SY↗

arXiv2022

Integrated Adaptive Control and Reference Governors for Constrained Systems with State-Dependent Uncertainties

This paper presents an adaptive reference governor (RG) framework for a linear system with matched nonlinear uncertainties that can depend on both time and states, subject to both state and input constraints. The proposed framework leverages an L1 adaptive controller (L1AC) that estimates and compensates for the uncertainties, and provides guaranteed transient performance, in terms of uniform bounds on the error between actual states and inputs and those of a nominal (i.e., uncertainty-free) system. The uniform performance bounds provided by the L1AC are used to tighten the pre-specified state and control constraints. A reference governor is then designed for the nominal system using the tightened constraints, and guarantees robust constraint satisfaction. Moreover, the conservatism introduced by the constraint tightening can be systematically reduced by tuning some parameters within the L1AC. Compared with existing solutions, the proposed adaptive RG framework can potentially yield less conservative results for constraint enforcement due to the removal of uncertainty propagation along a prediction horizon, and improved tracking performance due to the inherent uncertainty compensation mechanism. Simulation results for a flight control example illustrate the efficacy of the proposed framework.

eess.SY↗

arXiv2022

CVFNet: Real-time 3D Object Detection by Learning Cross View Features

In recent years 3D object detection from LiDAR point clouds has made great progress thanks to the development of deep learning technologies. Although voxel or point based methods are popular in 3D object detection, they usually involve time-consuming operations such as 3D convolutions on voxels or ball query among points, making the resulting network inappropriate for time critical applications. On the other hand, 2D view-based methods feature high computing efficiency while usually obtaining inferior performance than the voxel or point based methods. In this work, we present a real-time view-based single stage 3D object detector, namely CVFNet to fulfill this task. To strengthen the cross-view feature learning under the condition of demanding efficiency, our framework extracts the features of different views and fuses them in an efficient progressive way. We first propose a novel Point-Range feature fusion module that deeply integrates point and range view features in multiple stages. Then, a special Slice Pillar is designed to well maintain the 3D geometry when transforming the obtained deep point-view features into bird's eye view. To better balance the ratio of samples, a sparse pillar detection head is presented to focus the detection on the nonempty grids. We conduct experiments on the popular KITTI and NuScenes benchmark, and state-of-the-art performances are achieved in terms of both accuracy and speed.

cs.CV↗

arXiv2022

Optimizing Nitrogen Management with Deep Reinforcement Learning and Crop Simulations

Nitrogen (N) management is critical to sustain soil fertility and crop production while minimizing the negative environmental impact, but is challenging to optimize. This paper proposes an intelligent N management system using deep reinforcement learning (RL) and crop simulations with Decision Support System for Agrotechnology Transfer (DSSAT). We first formulate the N management problem as an RL problem. We then train management policies with deep Q-network and soft actor-critic algorithms, and the Gym-DSSAT interface that allows for daily interactions between the simulated crop environment and RL agents. According to the experiments on the maize crop in both Iowa and Florida in the US, our RL-trained policies outperform previous empirical methods by achieving higher or similar yield while using less fertilizers

cs.LG↗

arXiv2022

Robustifying Reinforcement Learning Policies with $\mathcal{L}_1$ Adaptive Control

A reinforcement learning (RL) policy trained in a nominal environment could fail in a new/perturbed environment due to the existence of dynamic variations. Existing robust methods try to obtain a fixed policy for all envisioned dynamic variation scenarios through robust or adversarial training. These methods could lead to conservative performance due to emphasis on the worst case, and often involve tedious modifications to the training environment. We propose an approach to robustifying a pre-trained non-robust RL policy with $\mathcal{L}_1$ adaptive control. Leveraging the capability of an $\mathcal{L}_1$ control law in the fast estimation of and active compensation for dynamic variations, our approach can significantly improve the robustness of an RL policy trained in a standard (i.e., non-robust) way, either in a simulator or in the real world. Numerical experiments are provided to validate the efficacy of the proposed approach.

cs.LG↗

arXiv2022

$\mathcal{L}_1$ Adaptive Augmentation for Geometric Tracking Control of Quadrotors

This paper introduces an $\mathcal{L}_1$ adaptive control augmentation for geometric tracking control of quadrotors. In the proposed design, the $\mathcal{L}_1$ augmentation handles nonlinear (time- and state-dependent) uncertainties in the quadrotor dynamics without assuming or enforcing parametric structures, while the baseline geometric controller achieves stabilization of the known nonlinear model of the system dynamics. The $\mathcal{L}_1$ augmentation applies to both the rotational and the translational dynamics. Experimental results demonstrate that the augmented geometric controller shows consistent and (on average five times) smaller trajectory tracking errors compared with the geometric controller alone when tested for different trajectories and under various types of uncertainties/disturbances.

eess.SY↗

arXiv2021

$\mathcal{L}_1$ Adaptive Control for Switching Reference Systems: Application to Flight Control

This paper presents a framework for the design and analysis of an $\mathcal{L}_1$ adaptive controller with a switching reference system. The use of a switching reference system allows the desired behavior to be scheduled across the operating envelope, which is often required in aerospace applications. The analysis uses a switched reference system that assumes perfect knowledge of uncertainties and uses a corresponding non-adaptive controller. Provided that this switched reference system is stable, it is shown that the closed-loop system with unknown parameters and disturbances and the $\mathcal{L}_1$ adaptive controller can behave arbitrarily close to this reference system. Simulations of the short period dynamics of a transport class aircraft during the approach phase illustrate the theoretical results.

math.OC↗

arXiv2020

Adaptive Robust Quadratic Programs using Control Lyapunov and Barrier Functions

This paper presents adaptive robust quadratic program (QP) based control using control Lyapunov and barrier functions for nonlinear systems subject to time-varying and state-dependent uncertainties. An adaptive estimation law is proposed to estimate the pointwise value of the uncertainties with pre-computable estimation error bounds. The estimated uncertainty and the error bounds are then used to formulate a robust QP, which ensures that the actual uncertain system will not violate the safety constraints defined by the control barrier function. Additionally, the accuracy of the uncertainty estimation can be systematically improved by reducing the estimation sampling time, leading subsequently to reduced conservatism of the formulated robust QP. The proposed approach is validated in simulations on an adaptive cruise control problem and through comparisons with existing approaches.

math.OC↗

arXiv2020

$\mathcal{L}_1$-$\mathcal{GP}$: $\mathcal{L}_1$ Adaptive Control with Bayesian Learning

We present $\mathcal{L}_1$-$\mathcal{GP}$, an architecture based on $\mathcal{L}_1$ adaptive control and Gaussian Process Regression (GPR) for safe simultaneous control and learning. On one hand, the $\mathcal{L}_1$ adaptive control provides stability and transient performance guarantees, which allows for GPR to efficiently and safely learn the uncertain dynamics. On the other hand, the learned dynamics can be conveniently incorporated into the $\mathcal{L}_1$ control architecture without sacrificing robustness and tracking performance. Subsequently, the learned dynamics can lead to less conservative designs for performance/robustness tradeoff. We illustrate the efficacy of the proposed architecture via numerical simulations.

eess.SY↗

arXiv2020

Novel Stealthy Attack and Defense Strategies for Networked Control Systems

This paper studies novel attack and defense strategies, based on a class of stealthy attacks, namely the zero-dynamics attack (ZDA), for multi-agent control systems. ZDA poses a formidable security challenge since its attack signal is hidden in the null-space of the state-space representation of the control system and hence it can evade conventional detection methods. An intuitive defense strategy builds on changing the aforementioned representation via switching through a set of carefully crafted topologies. In this paper, we propose realistic ZDA variations where the attacker is aware of this topology-switching strategy, and hence employs the following policies to avoid detection: (i) pause, update and resume ZDA according to the knowledge of switching topologies; (ii) cooperate with a concurrent stealthy topology attack that alters network topology at switching times, such that the original ZDA is feasible under the corrupted topology. We first systematically study the proposed ZDA variations, and then develop defense strategies against them under the realistic assumption that the defender has no knowledge of attack starting, pausing, and resuming times and the number of misbehaving agents. Particularly, we characterize conditions for detectability of the proposed ZDA variations, in terms of the network topologies to be maintained, the set of agents to be monitored, and the measurements of the monitored agents that should be extracted, while simultaneously preserving the privacy of the states of the non-monitored agents. We then propose an attack detection algorithm based on the Luenberger observer, using the characterized detectability conditions. We provide numerical simulation results to demonstrate our theoretical findings.

eess.SY↗

arXiv2019

Detectability of Intermittent Zero-Dynamics Attack in Networked Control Systems

This paper analyzes stealthy attacks, particularly the zero-dynamics attack (ZDA) in networked control systems. ZDA hides the attack signal in the null-space of the state-space representation of the control system and hence it cannot be detected via conventional detection methods. A natural defense strategy builds on changing the null-space via switching through a set of topologies. In this paper, we propose a realistic ZDA variation where the attacker is aware of this topology-switching strategy, and hence employs the policy to avoid detection: "pause (update and resume) attack" before (after) topology switching to evade detection. We first systematically study the proposed ZDA variation, and then develop defense strategies under the realistic assumptions. Particularly, we characterize conditions for detectability of the proposed ZDA variation, in terms of the network topologies to be maintained, the set of agents to be monitored, and the measurements of the monitored agents that should be extracted. We provide numerical results that demonstrate our theoretical findings.

eess.SY↗