A5015770407- Reinforcement Learning in Robotics
- Video Surveillance and Tracking Methods
- Anomaly Detection Techniques and Applications
- Robot Manipulation and Learning
- Autonomous Vehicle Technology and Safety
- Evacuation and Crowd Dynamics
- Visual Attention and Saliency Detection
- Context-Aware Activity Recognition Systems
- Virtual Reality Applications and Impacts
- Multimodal Machine Learning Applications
- Teleoperation and Haptic Systems
- Advanced Image and Video Retrieval Techniques
- Robotics and Sensor-Based Localization
- Robotics and Automated Systems
- Visual perception and processing mechanisms
- Robotic Path Planning Algorithms
- Urban Green Space and Health
- Human Pose and Action Recognition
Karlsruhe Institute of Technology
2024
Harbin Institute of Technology
2019-2021
This work explores how force feedback affects various aspects of robot data collection within the Extended Reality (XR) setting. Force has been proved to enhance user experience in by providing contact-rich information. However, its impact on not received much attention robotics community. paper addresses this shortcoming conducting an extensive study effects during XR. We extended two XR-based control interfaces, Kinesthetic Teaching and Motion Controllers, with haptic features. The is...
This paper introduces IRIS, an immersive Robot Interaction System leveraging Extended Reality (XR), designed for robot data collection and interaction across multiple simulators, benchmarks, real-world scenarios. While existing XR-based systems provide efficient intuitive solutions large-scale collection, they are often challenging to reproduce reuse. limitation arises because current highly tailored simulator-specific use cases environments. IRIS is a novel, easily extendable framework that...
Future versatile robots need the ability to learn new tasks and behaviors from demonstrations. Recent advances in virtual augmented reality position these technologies as great candidates for efficient intuitive collection of large sets While there are different possible approaches control a robot has not yet been an evaluation interfaces regards their efficiency intuitiveness. These characteristics become particularly important when working with non-expert users complex manipulation tasks....
Imitation learning with human data has demonstrated remarkable success in teaching robots a wide range of skills. However, the inherent diversity behavior leads to emergence multi-modal distributions, thereby presenting formidable challenge for existing imitation algorithms. Quantifying model's capacity capture and replicate this effectively is still an open problem. In work, we introduce simulation benchmark environments corresponding Datasets Diverse Demonstrations Learning (D3IL),...
The general environmental factors that influence fixation distribution as part of pedestrian visual behaviour under natural conditions are unclear. Relative luminance and saliency considered the parameters for predicting image-based fixation; however, they not confirmed by evidence from walking scenario. Field experiments using mobile eye-tracking glasses device were conducted on 16 participants in four commercial streets during day night. Fixation data along with processed images extracted...
In public scenes such as stations and hospitals, the crowds are intensive abnormal pedestrian often causes group hazards. The recognition of is an important security problem, which generally solved by inspection robots. Traditional visual feature methods pay much attention to inherent attributes pedestrians (such gender age), ignores complex semantic information displayed trajectories. This article uses scene monitoring sensors analyze trajectories in scenes. We propose trajectory framework,...
One of the challenging problems in robot navigation is efficient and safe planning a highly dynamic environment, where required to understand pedestrian patterns such as train station. The rapid movement pedestrians makes more difficult solve collision problem. In this paper, we propose probability map pedestrians’ problem based on influencer recognition model (IRM). (IRM) data-driven infer distribution over possible causes pedestrian’s turning. With model, can obtain by analyzing changes...
This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning (TOP-ERL), a novel algorithm that enables off-policy updates in the ERL framework. In ERL, policies predict entire action trajectories over multiple time steps instead of single actions at every step. These are typically parameterized by trajectory generators such as Movement Primitives (MP), allowing for smooth and efficient exploration long horizons while capturing high-level temporal correlations. However,...