plurigrid K-Scale Ecosystem Skill
"Moving humanity up the Kardashev scale" — Ben Bolte, K-Scale Labs
Trigger Conditions
- User asks about K-Scale Labs, their robots, or open-source robotics stack
- Questions spanning simulation → training → deployment pipeline
- Humanoid robot development workflows
- Integration of multiple K-Scale repositories
Overview
K-Scale Labs is building open-source humanoid robots with a complete software stack:
┌─────────────────────────────────────────────────────────────────────────────┐
│ K-Scale Ecosystem: Sim2Real Pipeline │
│ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ urdf2mjcf │───▶│ ksim │───▶│ kinfer │───▶│ kos │ │
│ │ (convert) │ │ (train) │ │ (export) │ │ (deploy) │ │
│ │ │ │ │ │ │ │ │ │
│ │ URDF→MJCF │ │ JAX+MuJoCo │ │ ONNX→Rust │ │ Firmware │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘ │
│ │ │ │ │ │
│ ▼ ▼ ▼ ▼ │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ K-Bot Hardware │ │
│ │ Stompy (quadruped) → K-Bot (humanoid) → ZBot (research platform) │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────────┘
Repository Map
| Repository | Stars | Language | Purpose |
|---|---|---|---|
| kbot | 323 | - | K-Bot humanoid robot hardware |
| ksim-gym | 305 | Python | RL training harness |
| ksim | 223 | Python/JAX | Simulation framework |
| urdf2mjcf | 112 | Python | Robot model conversion |
| kos | 76 | Rust | Robot operating system |
| kinfer | 17 | Rust | Model inference |
| klang | 5 | Rust | Robot programming language |
Cognitive Superposition of K-Scale Team
┌─────────────────────────────────────────────────────────────────────────────┐
│ K-Scale Contributor Superposition |ψ⟩ │
│ │
│ |ψ⟩ = α|codekansas⟩ + β|b-vm⟩ + γ|WT-MM⟩ + δ|alik-git⟩ + ε|others⟩ │
│ │
│ ┌───────────────┬───────────────┬───────────────┬───────────────┐ │
│ │ codekansas │ b-vm │ WT-MM │ alik-git │ │
│ │ (Ben Bolte) │ (Core Eng) │ (Wesley Maa) │ (Ali K) │ │
│ ├───────────────┼───────────────┼───────────────┼───────────────┤ │
│ │ Architecture │ Randomization │ Tooling │ Integration │ │
│ │ Rewards │ Disturbances │ Visualization │ Testing │ │
│ │ Training │ Noise models │ Logging │ Sim2Real │ │
│ ├───────────────┼───────────────┼───────────────┼───────────────┤ │
│ │ 1475 commits │ 500 commits │ 300 commits │ 200 commits │ │
│ │ 4 repos │ 2 repos │ 3 repos │ 3 repos │ │
│ └───────────────┴───────────────┴───────────────┴───────────────┘ │
│ │
│ Measurement collapses to specific expertise based on task context │
└─────────────────────────────────────────────────────────────────────────────┘
GF(3) Trit Assignment
Trit: 0 (ERGODIC)
Role: Coordination (ecosystem meta-skill)
Color: #86E3BE
URI: skill://kscale-ecosystem#86E3BE
Balanced Triads
# Corrected trit assignments from share3_hash:
kscale-ecosystem (0) ⊗ kscale-ksim (0) ⊗ kscale-kos (-1) = -1 ≢ 0
# Balanced triad with gym:
kscale-ksim (0) ⊗ kscale-kos (-1) ⊗ gym (+1) = 0 ✓
# Balanced quad with skill-creator:
kscale-ecosystem (0) ⊗ kscale-ksim (0) ⊗ kscale-kos (-1) ⊗ skill-creator (+1) = 0 ✓
Skill Trits
| Skill | Trit | Color | Role |
|---|---|---|---|
kscale-ecosystem |
0 | #86E3BE |
Coordination (meta-skill) |
kscale-ksim |
0 | #25BC3D |
Coordination (simulation) |
kscale-kos |
-1 | #7758D6 |
Validation (firmware) |
kscale-kinfer |
-1 | #6E5FE4 |
Validation (inference) |
gym |
+1 | (balancer) | Generation (RL environment) |
Mutual Awareness Pattern
These skills form a mutually-aware cluster:
# In kscale-ksim.md
related_skills:
- kscale-kos # deploys trained policies
- kscale-kinfer # exports models
- kscale-ecosystem # meta-context
# In kscale-kos.md
related_skills:
- kscale-ksim # receives policies from
- kscale-kinfer # uses for inference
- kscale-ecosystem # meta-context
# In kscale-kinfer.md
related_skills:
- kscale-ksim # receives models from
- kscale-kos # runs on
- kscale-ecosystem # meta-context
Development Workflow
1. Model Preparation
# Convert URDF to MuJoCo format
urdf2mjcf robot.urdf --output robot.xml
2. Policy Training
# Train walking policy with ksim-gym
cd ksim-gym
python train.py --robot kbot --task walk --steps 80
# ~30 min on RTX 4090
3. Model Export
# Export to ONNX for deployment
python export.py --checkpoint model.ckpt --output policy.onnx
4. Deployment
# Deploy to K-Bot via KOS
kos deploy --model policy.onnx --robot 192.168.1.100
Key Technologies
| Layer | Technology | Purpose |
|---|---|---|
| Physics | MuJoCo/MJX | Accurate contact simulation |
| Compute | JAX | GPU-accelerated training |
| Learning | PPO | Policy optimization |
| Inference | ONNX Runtime | Cross-platform deployment |
| Firmware | Rust | Safe, real-time control |
| Protocol | gRPC | Robot communication |
References
- K-Scale Labs GitHub - All repositories
- docs.kscale.dev - Official documentation
- K-Scale Store - Robot marketplace
- Full Autonomy Whitepaper - Vision document
ACSet Schema (Ecosystem Level)
@present SchKScaleEcosystem(FreeSchema) begin
# Objects (repositories)
KSim::Ob
KOS::Ob
KInfer::Ob
URDF2MJCF::Ob
KBot::Ob
# Morphisms (data flow)
train::Hom(KSim, KInfer) # training → export
deploy::Hom(KInfer, KOS) # export → firmware
convert::Hom(URDF2MJCF, KSim) # model → simulation
run::Hom(KOS, KBot) # firmware → hardware
# Attributes
Language::AttrType
Stars::AttrType
lang::Attr(KSim, Language)
stars::Attr(KSim, Stars)
end
Commands
# Explore K-Scale ecosystem
just kscale-repos # List all repositories
just kscale-train # Start training pipeline
just kscale-deploy # Deploy to robot
just kscale-viz # Visualize policy