You are browsing as a guest. Sign up (or log in) to start making projects!

Mihai

@Mihai

Joined May 31st, 2026

  • 30Devlogs
  • 2Projects
  • 1Ships
  • 15Votes
Open comments for this post
Reposted by @Mihai

2h 44m 8s logged

The map is done! I added some bariers with a simple shader, but I might change them soon, as I dont like the way they look. I also added a new car for the race.

I’ve also done some optimzing, enabled mesh batching, and after i’m 100% done with the map I will also bake the lighting for even better performance.

I’ve decided to leave out traffic for now, as its’ pretty difficult to implement and I don’t have the time, since school is starting.

As a small to do list for the ship:
-Main menu with settings
-Start sequence
-Race Timer
-Ending Cutscene
-Model credit list
-Sounds
-Maybe some music

3
1
204
Open comments for this post
Reposted by @Mihai

2h 51m 16s logged

Soo, I made the car more believable. It has working brake lights (WOOW), and new engine sounds that make it sound more powerful, but also added something extra spice, like exhaust backfire, that is programically made, with a multitude of firecracker sounds that I then processed. Besides that I also added a turbo blow off sound. They use the simulated rpm to decide their pitch and intensity, while also considering the throttle difference and position.

Last thing (visible in the last seconds of the last video), I improved the crash physics with detachable parts and deformable chassis.

0
1
3
Open comments for this post

2h 51m 16s logged

Soo, I made the car more believable. It has working brake lights (WOOW), and new engine sounds that make it sound more powerful, but also added something extra spice, like exhaust backfire, that is programically made, with a multitude of firecracker sounds that I then processed. Besides that I also added a turbo blow off sound. They use the simulated rpm to decide their pitch and intensity, while also considering the throttle difference and position.

Last thing (visible in the last seconds of the last video), I improved the crash physics with detachable parts and deformable chassis.

0
1
3
Open comments for this post

2h 44m 8s logged

The map is done! I added some bariers with a simple shader, but I might change them soon, as I dont like the way they look. I also added a new car for the race.

I’ve also done some optimzing, enabled mesh batching, and after i’m 100% done with the map I will also bake the lighting for even better performance.

I’ve decided to leave out traffic for now, as its’ pretty difficult to implement and I don’t have the time, since school is starting.

As a small to do list for the ship:
-Main menu with settings
-Start sequence
-Race Timer
-Ending Cutscene
-Model credit list
-Sounds
-Maybe some music

3
1
204
Open comments for this post

3h 20m logged

I have been working on the map ever since the last devlog. There isn’t much to say as there is to show. The scene has gotten laggy with lag spikes that seem random, so I will have to investigate and see how to optimize the game.

0
0
23
Open comments for this post

3h 32m 8s logged

Alright, i’ve been busy collecting assets and making materials and collections in order to start work on the map, which I did, I started with the dock area, since I dont yet have many house models, and well, I think it’s looking pretty cool, ngl.

Next up I’ll go around the track in order to detail the section most. I am using the darkness and distance as a tool to be lazy, but also optimize. No point in detailing a part the player wont see. Also in my toolset is the fact the player will be moving fast and concentrating on the road, so it won’t be necessary for me to detail every inch of the map, instead focusing on intersections and areas the player will be slow enough to notice such details.

0
0
17
Open comments for this post

2h 10m 1s logged

The really fun part has begun! Im stylizing the game. Yeah, I decided to skip traffic for now and instead get busy making my environment. First I started by making a cool night sky and lighting, i turned on some settings, and I am pretty afraid that this game wont run decent at all when Im done, but ill do optimizing later, im sure it will be fine.

Im still in search of models, as modelling everything by myself is out of the scope of this little project. But it’s gonna be pretty cool looking, i believe.

0
0
5
Open comments for this post

1h 24m logged

Ok, i’ve worked around a bit, and decided that for the first ship I want to release ONE very good feeling race. So I wont complicate myself with tuning and other systems, it will be one car, one race. So the atmosphere will be very important. I already have the layout of the race done, it’s supposed to go trough the docks > coast > downtown > highway and then the race ends.
Also I believe making this small but very polished game will lend me more experience than something that will be overly complex, as many of my game projects end up.

I will start with the terrain an polishing the track to make it more interesting in terms of driving and then i’ll get started on the terrain and dock area

0
0
8
Open comments for this post

5h 4m 43s logged

The racing AI is done. I also made a small race manager which takes the cars, aligns them in a grid and then starts a countdown. It also keeps trak of the position and lap-times, but there is no UI for that yet.

I also started working on the Map for the demo, i’m still debating how big the demo will be, but probably it’ll be a few races. After I have a small map layout, i’ll get started on the traffic AI, and move on from there.

Im still trying to get the hang of this road tool I am using, and also getting the road dimensions right, but we’ll see.

0
0
16
Open comments for this post

3h 8m 25s logged

I’m very happy with the racing AI, while it’s not perfect and cannot avoid everything, like a wall in the middle of the road, I dont think that will be the case like ever. It can avoid other cars, either static or not and it uses a basic prediction to determine if it’s on a collision course with another car and act accordingly.

Also, for the path, as you could see from the beggining of the video, the path was already backed based on the road. The car then only decides when to brake and steer based on that path.

Next up i’ll make a demo race with multiple racers and position marking and stuff and then get started on the map.

0
0
12
Open comments for this post

5h 38m 18s logged

Alrighty, I decided to actually use my brain and came to the conclusion to instead bake the racing paths, instead of just having them create one at runtime. I made a small unity tool, I just place the checkpoints, click bake and a racing spline gets generated. If you look real close, you can see that the line follows an outside-inside-outside path for steering. Now that I have this, I will work on the racer Ai itself. First I will make it follow the track with optimal speed and then teach it to avoid obstacles, other cars and then even overtake.

0
0
18
Open comments for this post

3h 16m 26s logged

Okk, I feel like i’m getting close. As you can see, the car now tries to build a racing line, this is all in real time. While its still not perfect, I feel like I am getting close.

There are still some bugs related to the pathing on the road that I will have to fix, but overall its very good.

The system works by identifying an apex. The geometrical apex is considered in this case the point which has the biggest angle difference compared to its neighbours. Now, after it identifies the apex, it grabs all points leading to the apex, and shifts them to the outside, before starting to shift them to the inside towards the apex, all using a sine wave based lerp in order to get a nice arc. While this system is good, I feel like the driver script that is responsible with making the car follow the points is kind of goofed, so i’ll look into that.

0
0
11
Open comments for this post

6h 35m 55s logged

Racing ai, thats all I’ve been doing. It works fine-ish, i feel like Im close but it still overshoots, but I have some ideas on how to fix it.

0
0
11
Ship

MYLO is a portable integrated neural network development IDE.

Currently, building neural networks for your projects is a hassle, you have to write code, change apps for different tools and they are resource hungry.

MYLO is here to change that!

MYLO is split between 3 components, all with a different purpose:

The Web interface

  • It handles the entire user experience, it’s made to be lightweight, as no actual ML processing is happening on it, thus making it easy to host and usable on many low end devices.
  • Its made using Next.js, and its the gateway to using the powerful tools MYLO provides, tools like:
  • Visual neural network editor, just like any other visual editor, inspired by unreal engine blueprints, you right click to add new nodes and connect them in order to build your own neural network.
  • Model Optimization, it uses a custom Neural-Genetic algorithm that applies mutations to the model, retrains it and then, the better performing models get their attributes passed on to the next generation and so on…
  • Model diagnosis, can either be run using a dummy input or have your model retrained on the dataset itself for better info
  • Dataset analasys tools, to see data distribution and outliers, together with per column information and encoding suggestions
  • Neural network visualization, with in depth info about each node
  • Ai assistant with a BYOK system, that can use tool calling to actually modify the model itself
  • Customizable interface, with window system similar to visual studio code
  • Model catalog where people can upload and share their models/projects
  • Agent connection system, that allows you to use agents running in your local network/machine

Now while these are the enabled and visible tools, also implemented are cloud tools that are currently disabled because it’s costly to host them:

  • An agent orchestrator that using docker spins up agent instances on demand and routes them to your app
  • Cloud storage system for projects and models

Behind the scenes there is a lot more going on, there is a full documentation document linked in the repo readMe, which I encourage you to read, even tho its in romanian, you can easily translate it. The only thing is it that it doesn’t include more info about the BYOK system and security updates that where made after the documentation was written.

The processing Agent

  • It handles all of the ML processing
  • Because of the way it’s implemented, it allows it to run on dedicated machines while the interface just connects to it
  • It features the implementation of the descriptor system, a way to represent models while allowing them to be easily editable, while also having transferable weights and parameters. It sits at the base of the entire editor and optimization system.
  • all of the features mentioned in the interface are run on the agent itself and not on the interface.
    For a more in depth analasys consult the documentation

Orchestrator

  • It handles the creation and destruction of the cloud instances, using a custom docker image and container handling, nothing fancy

The biggest challenge…

was the descriptor system, as an app made only by me and my friend, it was very hard to implement a robust system, and for the moment it’s quite unfeasable to extend the system further, as currently for each high level node its necessary to create connection templates manually, so thats why the node selection is currently smaller. Also it has to be easily compiled and almost ready for runtime while still being maleable enough for editing.

Thankfully all of these hard systems I was able to build with help of a friend who is much more informed when it comes to AI.

For any issues please write on my email ([email protected]) or on slack (altough im not so active)

  • 18 devlogs
  • 62h
  • 16.00x multiplier
  • 994 Stardust
Try project → See source code →
Open comments for this post

3h 16m 27s logged

The app is now in a state ready for shipment, small UI tweaks and the possibility to easily enable and disable cloud features has been added.

I borrowed an powerful AI agent from a friend to run a security audit on my website and also used it to fix them, as to assure the app is secure.

0
0
22
Open comments for this post

2h 24m 57s logged

In order to prepare my app for the ship, I decided to add some documentation and instructions about how to use the app, so there is now a docs page. Also necessary for the ship and publication, I made a Bring Your Own Key system for the AI Assistant, this way I don’t have to maintain and risk any of my own API keys. The keys are encrypted and stored in a Database, and once in said database, they only ever reach the Next.js server, no browser.

The last few things I have to do for the ship are the following:
-Make it easy to enable/disable cloud features (For the initial ship there will be no cloud features as I can’t pay for any uptime)
-Consistent styling for the entire UI, fix any UI quirks I find, and/or redesign sections of the app.
-Compile and make a release version of the desktop agent.

0
0
19
Open comments for this post

4h 59m 17s logged

I’ve made some polishing and steps towards the ship, that I want to make it happen this week.

There have been a lot of bugs fixes and the addition of a new tool: The model diagnostic and retrain tool, which as it says, it helps you get metrics about a model on a specific dataset or, using a dummy input, get characteristics about the model itself.

Besides that I got started on a landing page and added some flare to the login page.

0
0
3
Open comments for this post

2h 4m 38s logged

AI agent guyz!!! It uses gemini and I gave it some tools to edit and understand the model. I also added a markdown visualiser to have nice and pretty outputs from the model. It wasnt hard to implement and it’s a great addition to the app I feel.

0
0
3
Open comments for this post

3h 15m 41s logged

I’ve added cloud project saving and a project catalog!
You can publish your projects/models for other to see, train and use in their own projects.

It uses a simple s3 bucket for storage and handling the models, nothing too complex.

The catalog has some filters and search function.

0
0
3
Open comments for this post

2h 36m 49s logged

I have added what I call an orchestrator system, you can now create a new client instance in the cloud, that can then be used for processing and editing your models, this comes as an automatic substitution to hosting your own local client.

From the home page you can now click on the add new configuration button and choose the cloud option and a automatic connection will be established.

The system uses docker to spin up and manage the instances, there is a service for cleanup and deleting orphan containers.

The orchestrator itself doesn’t act like a proxy, all of the request go directly between the next.js server of the interface and instance.

0
0
3
Loading more…

Followers

Loading…