A5014381337- Robotic Path Planning Algorithms
- Autonomous Vehicle Technology and Safety
- Adversarial Robustness in Machine Learning
- Fault Detection and Control Systems
- Gaussian Processes and Bayesian Inference
- Reinforcement Learning in Robotics
- Video Surveillance and Tracking Methods
- Target Tracking and Data Fusion in Sensor Networks
- Distributed Control Multi-Agent Systems
- Simulation Techniques and Applications
- Anomaly Detection Techniques and Applications
- Advanced Control Systems Optimization
- Maritime Navigation and Safety
- Data Management and Algorithms
- Traffic Prediction and Management Techniques
- Bayesian Modeling and Causal Inference
- Advanced Optimization Algorithms Research
- Underwater Vehicles and Communication Systems
- Domain Adaptation and Few-Shot Learning
- Control Systems and Identification
- Hydraulic and Pneumatic Systems
- Fuel Cells and Related Materials
- Advanced Neural Network Applications
- Formal Methods in Verification
- Matrix Theory and Algorithms
Intelligent Systems Research (United States)
2024
Stanford University
2022-2024
American Institute of Aeronautics and Astronautics
2023
Massachusetts Institute of Technology
2023
Engineering Systems (United States)
2020-2022
University of Virginia
2017-2022
Vaughn College of Aeronautics and Technology
2022
Institute of Electrical and Electronics Engineers
2020
Gorgias Press (United States)
2020
Vrije Universiteit Brussel
2020
As neural networks (NNs) become more prevalent in safety-critical applications such as control of vehicles, there is a growing need to certify that systems with NN components are safe. This paper presents set backward reachability approaches for safety certification feedback loops (NFLs), i.e., closed-loop policies. While strategies have been developed without components, the nonlinearities activation functions and general noninvertibility weight matrices make NFLs challenging problem. To...
Modern unmanned aerial vehicles (UAVs) rely on constant periodic sensor measurements to detect and avoid obstacles. However, checking replanning are time energy consuming often not necessary especially in situations which the UAV can safely fly uncluttered environments without entering unsafe states. Thus, this paper, we propose a self-triggered framework that leverages reachability analysis schedule next check perform while guaranteeing safety under noise disturbance effects. Further, relax...
Modern unmanned aerial vehicles (UAVs) rely on constant periodic sensor measurements to help avoid obstacles, deal with system disturbances and correct uncertainties. However, checking is time energy consuming often not necessary especially in situations which the UAV can safely fly using open loop control without entering unsafe states. Thus, this paper we focus two main problems: i) how predict possible reachable states of considering noise ii) when schedule next state monitoring...
Autonomous systems operating in uncertain environments under the effects of disturbances and noises can reach unsafe states even while using finetuned controllers precise sensors actuators. To provide safety guarantees on such during motion planning operations, reachability analysis (RA) has been demonstrated to be a powerful tool. RA, however, suffers from computational complexity, especially when dealing with intricate characterized by high-order dynamics, making it hard deploy for runtime...
Detection and segmentation of moving obstacles, along with prediction the future occupancy states local environment, are essential for autonomous vehicles to proactively make safe informed decisions. In this paper, we propose a framework that integrates two capabilities together using deep neural network architectures. Our method first detects segments objects in scene, uses information predict spatiotemporal evolution environment around vehicles. address problem direct integration both...
Due to changes in model dynamics or unexpected disturbances, an autonomous robotic system may experience unforeseen challenges during real-world operations which affect its safety and intended behavior: particular actuator failures external disturbances are among the most common causes of degraded mode operation. To deal with this problem, work, we present a meta-learning-based approach improve trajectory tracking performance unmanned aerial vehicle (UAV) under faults have not been...
Autonomous vehicles are typically developed and trained to work under certain system environmental conditions defined at design time can fail or perform poorly if unforeseen such as disturbances changes in model dynamics appear runtime. In this work, we present a fast online planning, learning, recovery approach for safe autonomous operations unknown runtime disturbances. Our estimates the behavior of with an provides plans previously unseen by leveraging Gaussian Process regression theory...
In this paper, we consider the problem of online planning and control heterogeneous robotic systems for energy constrained safety operations. Specifically, a team aerial ground vehicles is tasked with completing set previously unknown goals within bounded region automatically deciding future recharging operations to avoid depleting throughout mission. We propose prioritized distributed virtual potential scheme spring-mass magnetic force behaviors in order plan coordinate motion so that...
In this paper, we present a fast run-time monitoring framework for safety assurance during autonomous system operations in uncertain environments. Modern unmanned vehicles rely on periodic sensor measurements motion planning and control. However, vehicle may not always be able to obtain its state information due various reasons such as failures, signal occlusions, communication problems. To guarantee the of these circumstances under presence disturbance noise, propose novel reachability...
Model-based planners for partially observable problems must accommodate both model uncertainty during planning and goal objective inference. However, model-based may be brittle under these types of because they rely on an exact tend to commit a single optimal behavior. Inspired by results in the model-free setting, we propose entropy-regularized planner problems. Entropy regularization promotes policy robustness inference encouraging policies no more committed action than necessary. We...
Changes in model dynamics due to factors like actuator faults, platform aging, and unexpected disturbances can challenge an autonomous robot during real-world operations affecting its intended behavior safety. Under such circumstances, it becomes critical improve tracking performance, predict future states of the system, replan maintain safety liveness conditions. In this letter, we propose a meta-learning-based framework learn system's their uncertainties under unforeseen untrained...
Safe and reliable autonomy solutions are a critical component of next-generation intelligent transportation systems. Autonomous vehicles in such systems must reason about complex dynamic driving scenes real time anticipate the behavior nearby drivers. Human is highly nuanced specific to individual traffic participants. For example, drivers might display cooperative or non-cooperative behaviors presence merging vehicles. These be estimated incorporated planning process for safe efficient...
This paper addresses a safe planning and control problem for mobile robots operating in communication- sensor-limited dynamic environments. In this case the cannot sense objects around them must instead rely on intermittent, external information about environment, as e.g., underwater applications. The challenge is that plan using only stale data, while accounting any noise data or uncertainty environment. To address we propose compositional technique which leverages neural networks to...
One of the bottlenecks training autonomous vehicle (AV) agents is variability environments. Since learning optimal policies for unseen environments often very costly and requires substantial data collection, it becomes computationally intractable to train agent on every possible environment or task AV may encounter. This paper introduces a zero-shot filtering approach interpolate learned past experiences generalize ones. We use an experience kernel correlate These correlations are then...
Perception systems operate as a subcomponent of the general autonomy stack, and perception system designers often need to optimize performance characteristics while maintaining safety with respect overall closed-loop system. For this reason, it is useful distill high-level requirements into component-level on In work, we focus efficiently determining sets safe given black-box simulator fully-integrated, We combine advantages common estimation techniques such Gaussian processes threshold...
One of the bottlenecks training autonomous vehicle (AV) agents is variability environments. Since learning optimal policies for unseen environments often very costly and requires substantial data collection, it becomes computationally intractable to train agent on every possible environment or task AV may encounter.This paper introduces a zero-shot filtering approach interpolate learned past experiences generalize ones. We use an experience kernel correlate These correlations are then...
For autonomous vehicles to proactively plan safe trajectories and make informed decisions, they must be able predict the future occupancy states of local environment. However, common issues with prediction include predictions where moving objects vanish or become blurred, particularly at longer time horizons. We propose an environment framework that incorporates semantics for prediction. Our method first semantically segments uses this information along spatiotemporal evolution validate our...
Perception systems operate as a subcomponent of the general autonomy stack, and perception system designers often need to optimize performance characteristics while maintaining safety with respect overall closed-loop system. For this reason, it is useful distill high-level requirements into component-level on In work, we focus efficiently determining sets safe given black-box simulator fully-integrated, We combine advantages common estimation techniques such Gaussian processes threshold...
Safe and reliable autonomy solutions are a critical component of next-generation intelligent transportation systems. Autonomous vehicles in such systems must reason about complex dynamic driving scenes real time anticipate the behavior nearby drivers. Human is highly nuanced specific to individual traffic participants. For example, drivers might display cooperative or non-cooperative behaviors presence merging vehicles. These be estimated incorporated planning process for safe efficient...
This paper addresses a safe planning and control problem for mobile robots operating in communication- sensor-limited dynamic environments. In this case the cannot sense objects around them must instead rely on intermittent, external information about environment, as e.g., underwater applications. The challenge is that plan using only stale data, while accounting any noise data or uncertainty environment. To address we propose compositional technique which leverages neural networks to...
Detection and segmentation of moving obstacles, along with prediction the future occupancy states local environment, are essential for autonomous vehicles to proactively make safe informed decisions. In this paper, we propose a framework that integrates two capabilities together using deep neural network architectures. Our method first detects segments objects in scene, uses information predict spatiotemporal evolution environment around vehicles. To address problem direct integration both...
As neural networks (NNs) become more prevalent in safety-critical applications such as control of vehicles, there is a growing need to certify that systems with NN components are safe. This paper presents set backward reachability approaches for safety certification feedback loops (NFLs), i.e., closed-loop policies. While strategies have been developed without components, the nonlinearities activation functions and general noninvertibility weight matrices make NFLs challenging problem. To...