A5088276691- Reinforcement Learning in Robotics
- Robotic Path Planning Algorithms
- AI-based Problem Solving and Planning
- Robotics and Sensor-Based Localization
- Multi-Agent Systems and Negotiation
- Robot Manipulation and Learning
- Modular Robots and Swarm Intelligence
- Machine Learning and Algorithms
- Logic, Reasoning, and Knowledge
- Robotics and Automated Systems
- Optimization and Search Problems
- Artificial Intelligence in Games
- Robotic Locomotion and Control
- Social Robot Interaction and HRI
- Advanced Image and Video Retrieval Techniques
- Auction Theory and Applications
- Time Series Analysis and Forecasting
- Adversarial Robustness in Machine Learning
- Semantic Web and Ontologies
- Explainable Artificial Intelligence (XAI)
- Anomaly Detection Techniques and Applications
- Advanced Vision and Imaging
- Stock Market Forecasting Methods
- Evolutionary Algorithms and Applications
- Multimodal Machine Learning Applications
Carnegie Mellon University
2014-2025
JPMorgan Chase & Co (United States)
2021-2024
Morgan Stanley (United States)
2020-2024
Universidad Carlos III de Madrid
2023
University of Coimbra
2023
New York Proton Center
2021
Hudson Institute
2020
New York Academy of Sciences
2020
John Wiley & Sons (United States)
2020
Albert Einstein College of Medicine
2020
Fast changing, increasingly complex, and diverse computing platforms pose central problems in scientific computing: How to achieve, with reasonable effort, portable optimal performance? We present SPIRAL, which considers this problem for the performance-critical domain of linear digital signal processing (DSP) transforms. For a specified transform, SPIRAL automatically generates high-performance code that is tuned given platform. formulates tuning as an optimization exploits domain-specific...
Vision systems employing region segmentation by color are crucial in real-time mobile robot applications. With careful attention to algorithm efficiency, fast image can be accomplished using commodity capture and CPU hardware. This paper describes a system capable of tracking several hundred regions up 32 colors at 30 Hz on general purpose The software consists of: novel implementation threshold classifier, merging form through connected components, separation sorting that gathers various...
Activity recognition is a key component for creating intelligent, multi-agent systems. Intrinsically, activity temporal classification problem. In this paper, we compare two models classification: hidden Markov (HMMs), which have long been applied to the problem, and conditional random fields (CRFs). CRFs are discriminative labeling sequences. They condition on entire observation sequence, avoids need independence assumptions between observations. Conditioning observations vastly expands set...
The sheer volume of data generated by depth cameras provides a challenge to process in real time, particular when used for indoor mobile robot localization and navigation. We introduce the Fast Sampling Plane Filtering (FSPF) algorithm reduce 3D point cloud sampling points from image, classifying local grouped sets as belonging planes (the “plane filtered” points) or that do not correspond within specified error margin “outlier” points). then based on an observation model down-projects plane...
Building upon previous work that demonstrates the effectiveness of WiFi localization information per se, in this paper we contribute a mobile robot autonomously navigates indoor environments using sensory data. We model world as signature map with geometric constraints and introduce continuous perceptual environment generated from discrete graph-based signal strength sampling. our algorithm which continuously uses to update location conjunction its odometry then briefly navigation approach...
Abstract Planning is a complex reasoning task that well suited for the study of improving performance and knowledge by learning, i.e. accumulation interpretation planning experience. PRODIGY an architecture integrates with multiple learning mechanisms. Learning occurs at planner's decision points integration in achieved via mutually interpretable structures. This article describes planner, briefly reports on several modules developed earlier along project, presents more detail two recently...
We present a new localization algorithm, called sensor resetting localization, which is an extension of Monte Carlo localization. The algorithm adds based re-sampling to when the robot lost. Sensor (SRL) robust modelling errors including unmodelled movements and systematic errors. It can be used in real time on systems with limited computational power. has been successfully autonomous legged robots Sony league robotic soccer competition RoboCup'99. results from demonstrating success...
We contribute Policy Reuse as a technique to improve reinforcement learning agent with guidance from past learned similar policies. Our method relies on using the policies probabilistic bias where faces three choices: exploitation of ongoing policy, exploration random unexplored actions, and introduce algorithm its major components: an strategy include new reuse bias, similarity function estimate respect one. provide empirical results demonstrating that improves performance over different...
We present Confidence-Based Autonomy (CBA), an interactive algorithm for policy learning from demonstration. The CBA consists of two components which take advantage the complimentary abilities humans and computer agents. first component, Confident Execution, enables agent to identify states in demonstration is required, request a human teacher learn based on acquired data. selects demonstrations measure action selection confidence, our results show that using Execution requires fewer than...
SPIRAL is a generator for libraries of fast software implementations linear signal processing transforms. These are adapted to the computing platform and can be re-optimized as hardware upgraded or replaced. This paper describes main components SPIRAL: mathematical framework that concisely transforms their algorithms; formula captures at algorithmic level degrees freedom in expressing particular transform; translator encapsulates compilation when translating specific algorithm into an actual...
We contribute an approach for interactive policy learning through expert demonstration that allows agent to actively request and effectively represent examples. In order address the inherent uncertainty of human demonstration, we as a set Gaussian mixture models (GMMs), where each model, with multiple components, corresponds single action. Incrementally received examples are used training data GMM set. then introduce our confident execution approach, which focuses on relevant parts domain by...
Several researchers, present authors included, envision personal mobile robot agents that can assist humans in their daily tasks. Despite many advances robotics, such still face limitations perception, cognition, and action capabilities. In this work, we propose a symbiotic interaction between to overcome the while allowing robots also help humans. We introduce visitor's companion agent, as natural task for interaction. The visitor lacks knowledge of environment but easily open door or read...