Robotics&Physics Engine

A slight taste of automation

Inspired by Reinforcement learning, trying some cool robotics examples with open-source physic engine {MoJuCu} or {Unity ML toolkit}.

Matlab Simulation could be another mainstream choice. Here is a example using Unity, ROS & Matlab for pick and drop robot hand {Link}.

Here is a youtube example training a cube to reach a ball. {YouTube}, which includes Unity ML-Agent Toolkit, tensorboard visualization, mass training, and heuristic control testing (manual control). In the series, car-driving examples mentioned a technique Imitation Learning {Medium Link} that relies on expert instruction to assist the Agent

Currently, as a beginner, I just choose the most prevalent implementations.

3 min physics

javascripts
webGUI
Got this cool guy in youtube recommendation.
Gonna take a good look later on.
github_pagehttps://matthias-research.github.io/pages/tenMinutePhysics/index.html

Lagrangian & Hamiltonian Mechanics

Thanks to Congyi. I think these idea of constant momentum may fits the fact rock fragmentations or secondary rockfalls. Also Lagrangian mechanics seems naturally establishes a second-order partial derivative equation. Currently reading this post {Link}.

Euler-Lagranigna Equation:

[latex] \delta S = \int\limits_{t_0}^{t_e}{L(q, \dot{q}, t) dt} = \int\limits_{t_0}^{t_e}{\partial{L}} [/latex]

\(\delta S = \int_{t_1}^{t_2} \left[ \frac{\partial L}{\partial q} \delta q + \frac{\partial L}{\partial \dot{q}} \delta \dot{q} \right] dt\)

\(\delta S = \int_{t_1}^{t_2} \left[ \frac{\partial L}{\partial q} – \frac{d}{dt} \left( \frac{\partial L}{\partial \dot{q}} \right) \right] \delta q \, dt.\)

Hamiltonian, swap \(\dot{q}\) with independent momentum \(\dot{p}\)

Example: particle in central force field

Hardwares

Lidar module and processor

Nazar shows a Livox Mid-70 with 905 nm NIR {Official Website}; By physics, it can reach 2-3mm surface penetration, active laser emission avoid surface scattering and foliage blockage.

Hard to mount on off-the-shelf drones; Need at least Orange Pi 5 to power it for edge computing. (Muat have Ethernet port for high-speed UDP); Itself cannot distinguish moisture or color so usually needs multi-sensor fusion.

The Orange Pi 5 is a Shenzhen-based affordable alternative for Raspberry Pi and even Jetson Nano, which supports edge computing and onboard deep inference.