Only Truth Matters
0%

Posted on Edited on In , ,

1. Basic

From chatGPT

  1. Find area with natural light
  2. Invest
  3. Sturdy tripod
  4. Artifical lighting
  5. Light Modifiers
  6. Background and surfaces
  7. Styling props
  8. Organize the workspace
  9. Learn and Practice
  10. Portfolio
  11. Network and Market

2. Artifical Light

Type:

  • Continuous light
  • Strobe light or flash
  • Speedlight

Color Temperature

  • 5000-5500

CRI:

  • Above 90 is desireable
Read more »

Posted on Edited on In , ,

Ref

A student can only participate the STEM OPT twice

Must purpuse their practical training through a employer that is enreolled in E-verify

they may use a start-up business so long as all regulatory requirements are met

Ref2

May work for more than one employer, must be related to degree program and must be enrolled in E-verify

2. Steps froim SBA

Ref

  1. Conduct market research

  2. check if it is profitable

  3. Write business plan. Roadmap for how to structure, how to grow

  4. Fund your busniess

  5. Raise enough money

  6. Pick a location

  7. Choose structure

  8. sole proprietorship

  9. partnership

  10. LLC (limited liablity company)

Read more »

Posted on Edited on In , , , ,

1. Intro

Basic Method

  • Graph-based: A*, D*
  • Anytime Sampling: RRT, RPM

BIT: combine graph and sampling

  • Use RGG to approximate the searched space

Prove

  • Completeness: can always find a solution from start to goal (if exist)
  • Almost-sure Asymptotic optimality

2. Unknown Term

RGG: random geometry graph

  • A type of mathematical graph
  • Node is randomly distributed in geometric space
Read more »

Posted on Edited on In , , ,

1. What

For a IMU, it can measure

  • linear acceleration
  • angular velocity

What to calibrate:

  • measurement bias
  • constant over time, independent of sonsor motion
  • due to manufacturing tolerance
  • scale factor
  • constant over time
  • due to tolerance
  • misalignments
  • drift rate
  • Noise

2. Basic Math

Ref

How accelerator works: MEMS

accelerometer model

The basic is: detect force

Read more »

Posted on Edited on In , , , ,

1. What

Collision Testing algorithm

  • Very general, works with any combination of convex shape
  • Fast

Useful in physics simulation and video games

2. Math Foundations

Define a shape

  • A set of points that are infinitely dense

Define the intersection: share at least one point, vector difference is 0

  • if a=b
  • then a-b=0

If there are infinite number of points, it fails

Read more »

Posted on Edited on In , , , ,

1. Goal

Avoid harm

General step:

  1. Create 3D model for robot (from URDF)
  2. Define workspace and trajectory (to detect the potential collision)
  3. Detection
  4. Avoidance
  5. Test

2. Detection

Separating Axis Theorem (SAT): Simple but computationally expensive

  • Pros:
  • Suitable for detecting collisions between convex polyhedra.
  • Can handle objects of arbitrary shapes and sizes.
  • Relatively easy to implement.
  • Cons:
  • May produce false negatives if the objects are not convex or have concavities.
  • Can be computationally expensive for complex shapes with many faces or vertices.

Gilbert-Johnson-Keerthi (GJK) algorithm:

  • Pros:
  • Can detect the intersection between any two convex objects.
  • Efficient and robust algorithm.
  • Can handle objects of arbitrary shapes and sizes.
  • Cons:
  • Can be more difficult to implement than some other methods.
  • May produce false negatives if the objects are not convex or have concavities.
  • Requires an initial guess for the closest points between the two shapes.

Minkowski Portal Refinement (MPR) algorithm: Non-convex, computation expensive

Read more »

Posted on Edited on In , , ,

1. Goal

Highlight some of the progress in 2022 and vision for 2023

2. Language Vision and Generative models

2.1. Language

Language model, although the simple objective is to predict the next word, the results are quite amazing

Chain of Thought prompting, help the language model follow logical chain of thought

Trained on multi-language model, it shows translation ability

Scaling give new power

img

2.2. Vision

Read more »

Posted on Edited on In , , ,

1. GPT3

三个重要能力

  • Text Generation, using prompt
  • In-Context learning
  • World knowledge, factual knowledge and common sense

来源：大量的数据

2. From GPT3 to ChatGPT

图片

不同的微调指令激发出了不同的能力，这些能力本来就已经有了

  • 上下文学习
  • 对话
  • 代码

3. Code-Davinci

通过指令微调，直接产物是能够响应人类指令

Read more »