A5017613080- Robotic Path Planning Algorithms
- Robotics and Sensor-Based Localization
- Reinforcement Learning in Robotics
- Robot Manipulation and Learning
- Underwater Vehicles and Communication Systems
- Prosthetics and Rehabilitation Robotics
- Robotic Locomotion and Control
- Advanced Vision and Imaging
- Neural Networks and Reservoir Computing
- Muscle activation and electromyography studies
- Robotics and Automated Systems
- Robotic Mechanisms and Dynamics
- Tactile and Sensory Interactions
- Animal Behavior and Reproduction
- Insect and Arachnid Ecology and Behavior
- Control and Dynamics of Mobile Robots
- Explainable Artificial Intelligence (XAI)
- Digital Games and Media
- Video Analysis and Summarization
- Advanced Sensor and Energy Harvesting Materials
- Sports Analytics and Performance
- Genetics and Physical Performance
- Imbalanced Data Classification Techniques
- Model-Driven Software Engineering Techniques
- Power Line Inspection Robots
Universidade Federal do Rio Grande
2021-2024
Res Publica (Norway)
2023
Universidade Federal de Santa Maria
2015-2022
National Taiwan Normal University
2019
Osaka University
2005-2009
Reinforcement Learning (RL) has presented an impressive performance in video games through raw pixel imaging and continuous control tasks. However, RL performs poorly with high-dimensional observations such as images. It is generally accepted that physical state-based policies laser sensor measurements give a more sample-efficient result than learning by pixels. This work presents new approach extracts information from depth map estimation to teach agent perform the mapless navigation of...
Previous works showed that Deep-RL can be applied to perform mapless navigation, including the medium transition of Hybrid Unmanned Aerial Underwater Vehicles (HUAUVs). This paper presents new approaches based on state-of-the-art actor-critic algorithms address navigation and problems for a HUAUV. We show double critic with Recurrent Neural Networks improves performance HUAUVs using solely range data relative localization. Our achieved better transitioning capabilities solid generalization...
This paper presents a study on cricket behavior using an interdisciplinary robot/insect mixed society setup. Field crickets of the species Gryllus bimaculatus were allowed to interact with micro-robots equipped decoys. allows stimulation insect behaviors that are usually difficult bring out evoke insects alone, allowing consistent behavioral research. We performed set experiments focused comparative dominant and subordinate male after dominance dispute is settled. From these we able collect...
This paper presents a novel compliant spring system designed to be attached conventional robotics servo motor, turning it into series elastic actuator (SEA). The is composed by only two mechanical parts: torsional polyurethane and round aluminum support for link attachment. spring, had its design derived from iterative FEM-based optimization process. A magnetometer based circuit used measure angular displacement communicate through RS485 bus protocol.
Image restoration and image enhancement are critical processing tasks since good quality is mandatory for many applications. We particularly interested in the of ill-exposed images. These effects caused by sensor limitation or optical arrangement. They prevent details scene from being adequately represented captured image. proposed a deep neural network model due to number uncontrolled variables that impact acquisition. The can converge representative training data, loss, optimization...
This work presents Dimitri, an open-software & open-hardware robot torso equipped with modular low cost compliant joints, built to allow advanced research on human-robot interaction, force-aware object manipulation and environment exploration. The has 13 DOFs: 4 DOFs series elastic actuators in each arm, 3 for roll, pitch yaw of the waist, 2 head pan tilt. Our main innovation is employment a polyurethane based spring system attached conventional robotics servo motors, turning them into...
Deep Reinforcement Learning (Deep-RL) techniques for motion control have been continuously used to deal with decision-making problems a wide variety of robots. Previous works showed that Deep-RL can be applied perform mapless navigation, including the medium transition Hybrid Unmanned Aerial Underwater Vehicles (HUAUVs). These are robots operate in both air and water media, future potential rescue tasks robotics. This paper presents new approaches based on state-of-the-art Double Critic...
Deep Reinforcement Learning (DRL) has emerged as a promising approach to enhancing motion control and decision-making through wide range of robotic applications. While prior research demonstrated the efficacy DRL algorithms in facilitating autonomous mapless navigation for aerial terrestrial mobile robots, these methods often grapple with poor generalization when faced unknown tasks environments. This paper explores impact Delayed Policy Updates (DPU) technique on fostering new situations,...
The inverse kinematics problem is generally very complex and many traditional solutions are targeted only to robots of certain specific topologies. iterative method based on the (pseudo) Jacobian matrix a well-known, proven reliable general approach that can be applied wide variety manipulators. However, it relies linearizations valid within tight neighborhood around current pose manipulator. This requires robot move at short steps, intensively recalculating its trajectory along way, making...