This project proposes using deep learning and deep reinforcement learning techniques for object detection, obstacle avoidance, and navigation of mobile robots. Deep convolutional neural networks like YOLO, R-CNN, and SSD will be used for object detection from camera images to inform obstacle avoidance. Reinforcement learning algorithms like DQN and DDPG will control the robot's wheels based on detections to navigate autonomously. The robot will be designed and simulated in Gazebo using ROS2.
This project proposes using deep learning and deep reinforcement learning techniques for object detection, obstacle avoidance, and navigation of mobile robots. Deep convolutional neural networks like YOLO, R-CNN, and SSD will be used for object detection from camera images to inform obstacle avoidance. Reinforcement learning algorithms like DQN and DDPG will control the robot's wheels based on detections to navigate autonomously. The robot will be designed and simulated in Gazebo using ROS2.
This project proposes using deep learning and deep reinforcement learning techniques for object detection, obstacle avoidance, and navigation of mobile robots. Deep convolutional neural networks like YOLO, R-CNN, and SSD will be used for object detection from camera images to inform obstacle avoidance. Reinforcement learning algorithms like DQN and DDPG will control the robot's wheels based on detections to navigate autonomously. The robot will be designed and simulated in Gazebo using ROS2.
And Deep Reinforcement Learning For Navigating Mobile Robots Nirban Das(20376014) Abstract This project proposes a methodology for detecting obstacles for mobile robots. The methods are based on the state of the art deep learning techniques that can detect various objects in the images. We will also need object localization to attain autonomous behaviour. Deep Convonvolution Neural Networks (CNN) based methods like YOLO(You Look Only Once), R-CNN(Region-Based CNN), FR-CNN, SSD(Single Shot MultiBox Detector) will be used and will be compared when the mobile robot is put in a real-time dynamic environment. The predictions will then be used for controlling the wheels of the robot. Based on the objects detected the robot will be using deep reinforcement learning techniques to control the mobile robot. Deep reinforcement methods like DQN(Deep Q- Network), DDPG(Deep Deterministic Policy Gradient), will be used to get the optimal parameters to move the wheels of the robot. The robots will be designed in gazebo environment and to run the robot we will use ROS2 for the robot. Proposed work ● Manuver and control mobile robot ● use the camera for perception of environment ● Use state of the art deep learning algorithms ● Object detection for obstacle avoidance ● Control the robot using reinforcement learning Proposed work ● Using Deep Learning Methods to detect objects which are possible obstacles – CNN based algorithms – YOLO – R-CNN – SSD Proposed work ● To control the Robot Wheels use Reinforcement learning algorithms – DQN – DDPG ● Use ROS2 and gazebo for simulations