Active Project
Multi-robot Exploration in Communication Restricted Environments
The main objective of this research is to propose, implement, evaluate, and deploy methods for multi-robot exploration with communication restrictions.
Concluded and/or Archived
A DRL Approach for Object Transportation in Complex Environments
Robots capable of transporting objects are suitable for many applications with societal and economic impact, such as waste retrieval, disposal, and object manipulation in space or the deep sea. However, formulating a coherent action plan is not trivial due to the size of the search space and the object’s physical properties. With the recent advances in Deep Reinforcement Learning (DRL), in this work, we propose, implement, and deploy value-based Deep Reinforcement Methods to tackle the determination of high-level actions that form robust strategies combined with a Probabilistic Roadmap (PRM) method for object transportation through complex environments. The solution was evaluated in a simulation environment and deployed into a real robot. Our results show that DRL can learn strategies effectively, and the robot was able to accomplish its task.
Navigation Methods for Mobile Robots
Three methods for path planning and reactive navigation were validated, PRM, RRT, and Potential Fields. It should be noted that the RRT (Rapidly-exploring Random Trees) selects the leaf (subgoal) that is closest to the final goal and then generates a new path. This allows real-time behavior as you see in the video. All methods were validated with Non-holonomic and Holonomic robots in CoppeliaSim.
Frontier Exploration With Reinforcement Learning
Several methods were integrated, which includes probabilistic roadmaps, c-space generations, and occupancy grid. It is currently being used to validate some hypotheses regarding reinforcement learning for frontier navigation. All methods were validated in Coppelia Simulator. https://youtu.be/AVbgz9mP8OU
Shepherding with Temporal Difference Learning
This research project focused on exploring basic shepherding algorithms and behaviors. Portrays rule-based shepherds, behavioral shepherds, and reinforcement learning-based single shepherds. The main objective is to deploy more behaviors through deep learning methods. https://youtu.be/KXBTYhxmSbs
Roguelike Flooding Engine for Reinforcement Learning Agents
Flooding engine to simulate natural hazard conditions in real-time. Through the simulated environment, I envision creating reinforcement learning agents able to relocate citizens, in case of critical failures of a deterministic path-finding algorithm. https://github.com/Ophien/RogueLike-Flooding-Engine-for-Reinforcement-Learning
ADAM, Real-time reinforcement learning for locomotion in continuous 3D environments without waypoints or grid maps
ADAM is a Unity3D demo character used to promote the game engine. Recently I am studying how to deploy real-time reinforcement learning methods for autonomous robots to navigate through an environment without prior knowledge. Furthermore, my experiments are conducted without the presence of waypoints or fixed paths in the agent that could use as a reference to navigate or a grid map. https://youtu.be/R7L16IDkC3c
https://github.com/Ophien/H-GM-Humanoid-Robot-Real-Time-Trajectory-Manipulator
Philip K. Dick the Robot experiments
Experiments with Philip k. Dick the robot regarding the integration of a decentralized AI for human-robot interaction with reasoning. It was used alongside Sophia for some demo projects.
H-GM, Humanoid Robot Real-Time Trajectory Manipulator
Study, proposal, and deployment of a 3D visualization system for the visualization, real-time manipulation, and GAIT study of humanoid robots. Deployed with OpenGL, the visualization system enables the real-time interaction between a studied humanoid robot and an external user and also the visualization of 3D GAIT trajectories. https://youtu.be/Ymxfv_at50k
https://github.com/Ophien/H-GM-Humanoid-Robot-Real-Time-Trajectory-Manipulator
Eidolon, Low-Cost Research Humanoid Robot
Study, proposal, and deployment of a low-cost humanoid robot research platform. I built this robot from scratch to conduct research in several fields regarding humanoid robots, from stabilization to gait performance. https://youtu.be/rBHkrd9r-Eg
https://github.com/Ophien/H-GM-Humanoid-Robot-Real-Time-Trajectory-Manipulator
Inverse Kinematics real-time pose mapping
Study, proposal, and deployment of low a cost robotic limbs using servo motors to perform real-time inverse kinematics through heuristics. Used the Forward and Backward Reaching Inverse Kinematics (FABRIK) and a Jacobian Transpose method to control mobile robots’ limbs in real-time. The deployed humanoid robot encompasses a center of mass stabilization system used to perform a walking GAIT. https://github.com/Ophien/H-GM-Humanoid-Robot-Real-Time-Trajectory-Manipulator
Mobile Robots Research Platform Treadmill
Study, proposal, and deployment of a low-cost humanoid robot treadmill research platform, composed of wood and a continuous current motor, to enable to conduct of low-cost research for the GAIT study of mobile robots.
Creative Agent Reasoning
The conducted research enabled an agent to learn, by using an unsupervised learning model called Fusion Architecture for Learning Cognition and Navigation, how to navigate through a randomized minefield without prior knowledge. Proposed theoretical aspects based on the ART theory and The Honing Theory to deploy creative behavior into autonomous agents.
Graphics Processing Unit Fuzzy ART Massive Neuron Storage System
Studied, proposed, and deployed a fuzzy ART Adaptive Neural Network into a Graphics Processing Unit (GPU) for fast computation of fuzzy ART-based neurons. The proposal was deployed to an autonomous agent, to play a digital collectible card game called Hearthstone, that performs its reasoning through an Adaptive Neural Network coded with CUDA on an Nvidia Tesla K20.
https://github.com/Ophien/FAL-ANN
https://github.com/Ophien/GPU-Adaptive-Resonance-Associative-Map
The Honing Theory Hearthstone AI
Proposed, developed, and deployed, a model to represent information according to the Honing Theory through a digital medium, create/adapt a graph search heuristic to perform the neuron activation process described by the theory, simplified and optimized the proposed model, and helped with the domain data set organization, helped verifying the integrity of the working data set and conduct all relying experiments. https://github.com/Ophien/Honing-Theory-Knowledge-Based-Systems