Aja - Motion AI and Animation
Named after the Yoruba Orisha Aja (forest guide), this domain provides comprehensive motion processing from video input to production-ready animation. It covers pose estimation, skeletal retargeting, motion quality assessment, and format conversion.
Architecture#
Apps (4)#
| App | Framework | Port | Purpose |
|---|---|---|---|
svc-motion-pipeline |
Fastify | 8090 | Pipeline orchestration, jobs, result delivery |
svc-motion-ai |
Hono | 3040 | AI motion processing (video-to-motion, retargeting, live capture) |
svc-reference-video |
-- | -- | Reference video library (ingestion, search, annotation, collaboration) |
cli |
Commander | -- | CLI tools (aja process, convert, inspect, debug, config, jobs, health) |
Libraries (38)#
Core Motion Processing (5):
| Library | Purpose |
|---|---|
@aja/motion-formats |
Multi-format support (BVH, FBX, GLB, USD, Alembic, C3D, TRC, ASF/AMC) |
@aja/motion-processing |
Filters, smoothing, interpolation |
@aja/motion-quality |
Quality metrics and assessment |
@aja/motion-validation |
Data validation and error detection |
@aja/motion-integration |
Integration layer |
Video Analysis (4):
| Library | Purpose |
|---|---|
@aja/pose-lifting |
2D to 3D pose lifting |
@aja/depth-sensing |
Monocular and stereo depth |
@aja/multi-view-reconstruction |
Multi-camera 3D reconstruction |
@aja/human-mesh-recovery |
SMPL/SMPL-X mesh recovery |
Skeletal and Avatar (6):
| Library | Purpose |
|---|---|
@aja/skeleton-mapping |
Skeleton hierarchy conversion |
@aja/optimization-ik |
Inverse kinematics solvers |
@aja/avatar-library |
Avatar definitions and presets |
@aja/avatar-integration |
Avatar binding and animation |
@aja/avatar-preview-ui |
React avatar preview components |
@aja/bone-mapping-ui |
Interactive bone mapping UI |
Retargeting and Blending (5):
| Library | Purpose |
|---|---|
@aja/semantic-retargeting |
Semantic motion retargeting |
@aja/proportional-adaptation |
Body proportion adaptation |
@aja/blend-shape-retargeting |
Facial blend shape retargeting |
@aja/neural-retargeting |
ML-based retargeting |
@aja/animation-blending |
Motion blending and transitions |
Domain Pipelines (2):
| Library | Purpose |
|---|---|
@aja/domain-motion-pipelines |
Pre-configured pipelines per domain |
@aja/fitness-animation |
Fitness-specific analysis |
Infrastructure (8):
| Library | Purpose |
|---|---|
@aja/motion-pipeline-sdk |
TypeScript SDK |
@aja/motion-pipeline-sdk-python |
Python SDK with Jupyter support |
@aja/batch-inference |
Batch processing |
@aja/video-chunking |
Video splitting for processing |
@aja/model-optimization |
Model compression |
@aja/pipeline-parallelism |
Distributed processing |
@aja/distributed-workers |
Worker coordination |
@aja/result-aggregation |
Result combining |
Security and Compliance (6):
| Library | Purpose |
|---|---|
@aja/privacy-protection |
Privacy-preserving processing |
@aja/consent-management |
Consent tracking |
@aja/content-security |
Security protocols |
@aja/content-watermarking |
Digital watermarking |
@aja/content-moderation |
Content filtering |
@aja/data-retention |
Retention policy enforcement |
Utility (1):
| Library | Purpose |
|---|---|
@aja/pipeline-cache |
LRU caching for pipeline stages |
Technology Stack#
| Layer | Technologies |
|---|---|
| SDKs | TypeScript (Node.js), Python (Jupyter/ML) |
| Motion | Custom parsers for 13+ animation formats |
| ML Models | Pose estimation, depth sensing, mesh recovery |
| Build | @nx/js:tsc |
| Testing | Vitest |
Processing Pipeline#
Aja's motion processing flows through five stages:
1. Input Sources
├── Video files → Pose Estimation
├── Motion Capture (C3D, TRC)
├── Animation files (BVH, FBX)
└── Live MoCap hardware
│
2. Video Analysis Layer
├── Pose Lifting (2D → 3D)
├── Depth Sensing
├── Multi-View Reconstruction
└── Human Mesh Recovery (SMPL-X)
│
3. Motion Processing Layer
├── Filtering and Smoothing
├── Validation
└── Quality Assessment
│
4. Retargeting Layer
├── Skeleton Mapping
├── Proportional Adaptation
├── Neural Retargeting
└── Animation Blending
│
5. Output Formats
├── BVH, FBX, GLB
└── Engine-native formats
Domain Pipelines#
Pre-configured pipelines optimized for specific use cases:
| Domain | Capabilities |
|---|---|
| Fitness | Exercise form analysis, rep counting, range of motion |
| Dance | Choreography capture, style transfer, sync |
| Sports | Performance analysis, technique breakdown |
| Medical | Gait analysis, rehabilitation, clinical assessment |
| Gaming | Real-time capture, gesture recognition |
| Film/VFX | High-fidelity capture, facial performance |
Quality Metrics#
@aja/motion-quality evaluates motion data across four dimensions:
- Temporal Smoothness - Jitter detection, acceleration consistency
- Anatomical Plausibility - Joint limits, bone lengths, self-penetration
- Ground Contact - Foot sliding, plane alignment
- Semantic Accuracy - Pose matching, action confidence
Supported Motion Formats#
@aja/motion-formats handles 13+ formats:
BVH, FBX, GLB/GLTF, USD/USDA, Alembic (ABC), C3D, TRC, ASF/AMC, Collada (DAE), MDD, PC2, OBJ sequences, custom JSON
Development#
# Run the motion pipeline service
pnpm nx dev @aja/svc-motion-pipeline
# Run tests
pnpm nx test @aja/motion-formats
# Build all Aja libraries
pnpm nx run-many --target=build --projects=tag:scope:aja
CI/CD#
Aja uses the core CI pipeline (ci.yml) and the general ECS deployment pipeline
(deploy-ecs.yml).
Related Documentation#
- Architecture - Detailed Aja domain architecture
- Monorepo Architecture - Monorepo overview
- Cross-Domain Integration