The character is driven by ArticulationBodies and learns to follow mocap animation while physically balancing itself. Instead of asking the neural network to generate the entire walk from scratch, I’m using the mocap pose as the reference and letting the agent learn the smaller corrections needed to stay upright, maintain momentum, place its feet correctly, and recover from instability.

The long-term goal is a fully physics-driven third-person controller that still has the quality and personality of mocap animation, but can naturally react to forces, uneven movement, collisions, and eventually player movement commands.

Still training and tuning it, but it’s been a really interesting mix of animation, reinforcement learning, and robotics-style control.

Upgrade your project's movement systems with my complete physics and animation bundle on itch.io. For a strictly limited time, you can get three of my premium assets at a massive discount.

Included in the Bundle

• Self-Balancing Active Ragdoll: Create dynamic, physically accurate character movements and realistic stumble recoveries.
• Self-Balancing Wheeled Ragdoll: Implement highly stable, physics-driven wheeled characters perfect for robots or unique vehicle mechanics.
• Procedural Legs Animator: Achieve smooth, terrain-adapting foot placement without relying on complex keyframed walk cycles.

Pricing Details

Normally, purchasing these three assets together would cost $39.99.

For a short time, the entire bundle is available for just $1.99.

Get the bundle here: https://itch.io/s/197326/frostpunchgames-bundle

It started as a personal project on Unity actually, while I was learning procedural animation. but I kept expanding it and eventually decided to package it for other developers and port it to Godot.

The system includes:

• Procedural two-bone leg IK
• Predictive foot placement
• Foot planting and slope alignment
• Forward, backward, sideways, and turning steps
• Heel-to-toe foot roll when walking forward
• Reversed foot roll when walking backward
• Step-synchronized hip movement
• Airborne leg posing
• Adjustable settings through the Inspector

feel free to grab it here: https://frostpunchgames.itch.io/procedural-legs-animator-for-godot

I wanted a system that could either completely replace the Animator or run additively on top of it to strictly lock feet to the ground based on physics and speed.

Here are the core mechanics I used to make it feel weighty and grounded instead of floaty:

• The Inverted Pendulum (Hip Dipping): The system constantly measures the horizontal distance between the planted feet and the hips. The wider the stride, the further the hips are procedurally pulled toward the floor. This forces the character to naturally "squat" into wide stances or fast runs.
• Contrapposto & Spine Flexibility: When the character banks into a sharp turn or leans into a sprint, rotating the hips forward looks robotic. To fix this, the system takes the hip's physical error offset and mathematically counter-rotates the chest and spine bones in the opposite direction. It keeps the head plumb and balanced over the center of mass.
• Velocity Trajectory: The IK solver doesn't just look at where the character is; it reads the smoothed velocity and angular velocity to predict where they will be. If you yank the joystick sideways, the character will step wide into the strafe to catch their balance before the center of gravity shifts too far.
• Terrain Alignment: It uses a custom Ground Probe to fire raycasts on the predicted step targets. The ankles dynamically calculate the ground normal and rotate to match the slope, so no more toes clipping into stairs or ramps.

I built a custom two-bone solver for the IK to keep it highly optimized, and I finally got around to building a clean custom inspector so I don't have to look at 50 different spring stiffness and damping variables anymore.

If anyone is struggling with foot sliding or wants to poke around the code to use it for their own bipeds, I packaged the whole system up and threw it on my itch page here: https://frostpunchgames.itch.io/procedural-legs-animator

it's an inverted-pendulum algorithm and active servos drives the wheels to catch the character's falling weight, creating natural momentum and flawless stability without kinematic cheating. I’d absolutely love to get this community's feedback on the physics feel, also, I've released it as early access on itch here if you want to grab it: https://frostpunchgames.itch.io/self-balancing-wheeled-ragdoll

Repost to another community

I’d absolutely love to get this community's feedback on the physics feel, also, I've released it as early access on itch here if you want to grab it: https://frostpunchgames.itch.io/self-balancing-wheeled-ragdoll

I’ve been feeling genuinely anxious about the future of our craft. The tools are evolving so fast that it’s hard not to wonder if the roles we’ve spent years mastering are about to become obsolete.

I recently ran an experiment that brought all these fears to the surface. I wanted to see exactly what happens when you use basic, greyscale 3D proxy models to render out a full 3D scene video with AI.

This technology is undeniably powerful, but it leaves me asking some hard questions about where we fit in the pipeline of tomorrow.
Would you use AI to render your scenes to cut time and cost, even if it disrupts the traditional roles on your team?

And more importantly, how do you feel as a 3D artist about AI rendering? Are you embracing it as the next evolution of our toolset, or does it feel like a direct threat to your livelihood?

Please share your thoughts below.

you can join the discord server for development updates, requests and release news:
https://discord.gg/5ZyqjTDhn4

Transform anything on your desktop with Screen Diffusion V0.2, an open-source, real-time AI generation tool.
https://github.com/rudyaa-sd/ScreenDiffusion

I'm trying to make an ML character controller, such a challenge to keep him upright!