Introduced by Lin et al. in Feature Pyramid Networks for Object Detection A Feature Pyramid Network, or FPN, is a feature extractor that takes a single-scale image of an arbitrary size as input, and outputs proportionally sized feature maps at multiple levels, in a fully convolutional fashion. This process is independent of the backbone convolutional architectures. It therefore acts as a generic solution for building feature pyramids inside deep convolutional networks to be used in tasks like object detection. The construction of the pyramid involves a bottom-up pathway and a top-down pathway. The bottom-up pathway is the feedforward computation of the backbone ConvNet, which computes a feature hierarchy consisting of feature maps at several scales with a scaling step of 2. For the feature pyramid, one pyramid level is defined for each stage. The output of the last layer of each stage is used as a reference set of feature maps. For ResNets we use the feature activations output by each stage’s last residual block. The top-down pathway hallucinates higher resolution features by upsampling spatially coarser, but semantically stronger, feature maps from higher pyramid levels. These features are then enhanced with features from the bottom-up pathway via lateral connections. Each lateral connection merges feature maps of the same spatial size from the bottom-up pathway and the top-down pathway. The bottom-up feature map is of lower-level semantics, but its activations are more accurately localized as it was subsampled fewer times. Self-Coding Performance

image image image image image

image image image image image image

Getting started %matplotlib inline import cv2 import random import numpy as np import matplotlib.pyplot as plt from google.colab.patches import cv2_imshow ]: %%capture ! wget -O img1.jpg "https://drive.google.com/uc?,→export=download&id=1omMydL6ADxq_vW5gl_1EFhdzT9kaMhUt" ! wget -O img2.jpg "https://drive.google.com/uc?,→export=download&id=12lxB1ArAlwGn97XgBgt-SFyjE7udMGvf"

Brief description image MCU Raspberry Pi 4B (RPi) ACTING Cascade Control & PID 3-DOF rigid-body coordinate transforms Planar kinematics of a differential-drive ground robot

Date: 202207 Problem Definition : Voice Conversion Deep Learning Modeling Datasets Methodology Overview Autoencoder • Ada-VC:

Three methods https://kknews.cc/zh-tw/news/3nkrkey.html AGV car traffic control regulation method based on path analysis http://www.agvbaike.com/agvyy/2019-09-25/6791.html workstations design https://www.sciencedirect.com/science/article/pii/S036083520800137X#tbl1 Agv zone strategies google:Agv zone strategies https://www.sciencedirect.com/science/article/pii/S0166361599000688#FIG2

:::success 2021 Intern @ Industrial Technology Research Institute (ITRI) :::

Mutitasks: Odometry: Fixed point moving VSLAM: Exploring unknown environments via VSLAM Object Detection: Detecting and grasping blocks with different colors and size. Demo Odometry VSLAM

Arduino & ROS file location: https://drive.google.com/drive/folders/139Qo4cpKLf0iU39zv5uE5DMH5dq13d3l?usp=sharing :::success Multiple sensor files, and versions for Arduino(.ino) and for ROS(C/C++) ::: Check steps Determine the threshold in the environment Place the car in the field about 80cm away from the light ball and turn it at an angle, about 700-800 degrees without light

Brief description Mutitasks: Gathered the state data from Armbot, and build DataLoader (State, Action, Next State) via Pytorch. Designed the Model Predictive Path Integral (MPPI) algorithm and loss function to estimate the target state. Demonstrated Armbot can avoid obstacles via the MPPI algorithm in an obstacle environment. Planar Pushing Learning State Space and Action Space For the planar pushing task, we have the following action and state spaces:

This paper develops a novel velocity estimator using an Kalman Filter (KF). The proposed state estimator for automotive vehicles combines data from an inertial measurement unit (IMU), wheel encoders, and GPS to produce a velocity estimate that is robust against wheel slip events and point out the section of slipping. The algorithm is efficient and can be implemented to run in real-time to produce accurate velocity estimates for vehicle control systems. Open source software is available for download and reproducing the presented results. Getting Started To run the Python code, the following environment and dependencies are required: Python == 3.6 numpy == 1.20.3 matplotlib == 3.4.3 scipy == 1.7.1

Code Source Gitlab

Brief description Mutitasks: Used OpenCV to detect blocks and implemented inverse kinematics to perform path planning of the manipulator. Calibrated camera using OpenCV and checkerboard to find the camera extrinsic and intrinsic matrix. Built a robot arm system that classifies and grasped a stack of blocks with different colors and size. Goal: Object Detection: Color and size can be distinguished Object Detection Steps

image image image image

Assignment introduction: The link Code organization in main folder ​​​​./root ​​​​├── README.pdf # Document explaining the implementation ​​​​├── README.md # Readme file for the workspace ​​​​├── planner.py # The code of the main Planner ​​​​├── solution.txt # Output of all paths of the king ​​​​├── problem-tests # The folder of testing data ​​​​│ ├──1 ​​​​│ │ ├── kings.txt # the start and goal position of kings

Leslie Chiu Ref:Build Scenes from Custom Data Using RoadRunner HD Map(C++ version) This example shows how to use RoadRunner HD Map protocol buffers to import custom data into RoadRunner and build scenes. In this example, you: Generate a RoadRunner HD Map file from custom data. Import the generated file into RoadRunner and preview the RoadRunner HD Map data. Build a RoadRunner scene from imported data. To build a scene, you must have a RoadRunner Scene Builder license. This example uses python to implement the RoadRunner HD Map using protocol buffers.

tage: MathWorks, perception Leslie Chiu image

image image image

[TOC] Final Project Report 簡介工作細節 :::success 我負責為XTx資料傳輸，XTx在衛星電機次系統功能為向基地台傳送資料，並接受EUCC指令，此6U衛星為傳送圖像資料為主，以下為XTx在本實作整體程序(mode2-3)。 註解： 本說明為 LOOPBACKMODE 因此未與EUCC連接，是STM32自傳自收 ::: 工作環境