Project Management
Auraison Sprint Plan - Sprint 16 onward
Generated: 2026-06-27 (regenerated from live Jira, AURA project, post-restructure) | Updated: 2026-06-27 Planning horizon: Sprints 16--21 (12 weeks, 2026-06-30 to 2026-09-19) Assignee: Pantelis Monogioudis (availability 0.6) Sprint capacity: 2 weeks x 5 days x 0.6 availability x 0.8 buffer = 4.8 days/sprint ~ 4 effective days
Source of truth is Jira
This plan is generated from the AURA project at aegean-ai.atlassian.net. Every row links to its
Jira issue. Regenerate after Jira changes via the Atlassian MCP:
project = AURA AND issuetype = Epic AND statusCategory != Done, then each epic's open children.
Scope
The June 2026 restructure consolidated the backlog to one governing epic per project, archived the legacy pre-2026 work under Legacy/Archive (AURA-214), and folded research threads into a single Research Book (AURA-716). This plan covers the 14 open platform and application epics. Out of scope: Legacy/Archive (AURA-214) and Voice & Speech Agents (AURA-211).
Summary
| Metric | Value |
|---|---|
| Open epics | 14 |
| Open child issues | 141 |
| In progress | AR4 sim (AURA-234), Ray Serve + vLLM (AURA-577), CF AI Gateway (AURA-705), Ray on Proxmox (AURA-507), VisDrone (AURA-559), artifacts/R2 (AURA-580), RoboMaker worlds (AURA-600/AURA-612) |
| Priority order | Infra → AR4 → Data plane → Control plane → Docs → Apps/Research |
| Sprints planned | 6 (Sprints 16--21) |
Epics
| Epic | Title | Status | Open |
|---|---|---|---|
| AURA-216 | AR4 Physical-AI: model-agnostic VLA control (LeRobot + GR00T) | In Progress | 18 |
| AURA-670 | Infrastructure: compute, storage, and deployment platform | To Do | 18 |
| AURA-176 | Data plane: lakehouse package (COCO-Caption Experiment #0) | To Do | 9 |
| AURA-544 | Aura: agent orchestrator (harness: Claude Agent SDK / Pi) | To Do | 9 |
| AURA-709 | Backend Algorithm Libraries & API (shared) | To Do | 2 |
| AURA-213 | Docs & Architecture | To Do | 33 |
| AURA-210 | User Plane: Robotics, VLA & Simulation | To Do | 10 |
| AURA-712 | TurtleBot Maze — autonomous navigation reference app | To Do | 6 |
| AURA-677 | CUA PDF Reader — agentic document understanding | To Do | 10 |
| AURA-688 | Deep Evidence Agents (DEA) | To Do | 0 |
| AURA-713 | Sidewalk Detection | To Do | 0 |
| AURA-714 | Tube Quality Control | To Do | 0 |
| AURA-715 | Counter-UAS | To Do | 0 |
| AURA-716 | Research Book | To Do | 26 |
Backlog by epic
Estimates (days) are indicative. Only the actively-scheduled epics are itemized below; the application epics and Research Book are summarized at the end.
Infrastructure (AURA-670) — P1
| Issue | Type | Title | Est | Status |
|---|---|---|---|---|
| AURA-507 | Task | Single-VM Ray cluster on Proxmox (native launcher) | 3 | In Progress |
| AURA-577 | Story | Ray Serve + vLLM model serving on gpu-node-3 | 5 | In Progress |
| AURA-705 | Task | Wire Cloudflare AI Gateway in front of local vLLM | 2 | In Progress |
| AURA-97 | Task | Launch distributed Ray cluster + test vLLM serving | 3 | To Do |
| AURA-346 | Task | Join GPU server as Ray worker (Docker + NVIDIA runtime) | 3 | To Do |
| AURA-518 | Task | Infra / Ray + Docker GPU — cluster, job service, containers | 5 | To Do |
| AURA-219 | Story | Policy Server: model-agnostic VLA inference endpoint | 5 | To Do |
| AURA-245 | Task | Standalone LeRobot policy inference container with Zenoh | 3 | To Do |
| AURA-704 | Task | Model Serving — vLLM / Ray Serve inference | 3 | To Do |
| AURA-337 | Story | Optimize Docker notebook execution to reuse venv | 2 | To Do |
| AURA-613 | Story | Adapt HF Model Trainer skill notebook for gpu-node-3 | 2 | To Do |
| AURA-614 | Story | Extend HF Model Trainer skill to local GPU execution | 2 | To Do |
| AURA-669 | Bug | Fix gpu-node-3 connection to Coolify server | 1 | To Do |
| AURA-328 | Bug | Add DMARC records for aegean.ai (email spoofing) | 1 | To Do |
| AURA-385 | Task | Create Claude Code Jira service account | 1 | To Do |
| AURA-400 | Task | Deploy Streamlit app to Hugging Face Spaces | 2 | To Do |
| AURA-671 | Task | Migrate files OneDrive → Google Drive via RiceDrive | 1 | To Do |
| AURA-703 | Task | Consolidate work credentials into 1Password "Work" vault | 1 | To Do |
AR4 Physical-AI (AURA-216) — P1 (In Progress)
| Issue | Type | Title | Est | Status |
|---|---|---|---|---|
| AURA-234 | Task | AR4 Gazebo pick-and-place simulation | 5 | In Progress |
| AURA-217 | Task | AR4 URDF + Gazebo Harmonic sim with gz_ros2_control | 5 | To Do |
| AURA-238 | Task | Record pick-and-place episodes via lerobot-ros | 3 | To Do |
| AURA-239 | Task | Implement MoveIt pick-and-place demo script | 3 | To Do |
| AURA-240 | Task | Create tabletop simulation launch file | 2 | To Do |
| AURA-218 | Task | MoveIt2 configuration for AR4 motion planning | 3 | To Do |
| AURA-220 | Task | Skill library: manipulation primitives for AR4 | 3 | To Do |
| AURA-221 | Task | Episode logging and LeRobot dataset builder | 3 | To Do |
| AURA-222 | Task | AR4 digital twin integration with data-plane | 3 | To Do |
| AURA-223 | Task | AR4 hardware bridge: Teensy 4.1 serial interface | 3 | To Do |
| AURA-224 | Story | GR00T N1 integration via Policy Server backend swap | 3 | To Do |
| AURA-225 | Task | Synthetic data pipeline via GR00T-Dreams and Cosmos | 3 | To Do |
| AURA-244 | Task | Build Zenoh-to-ROS action bridge for policy commands | 3 | To Do |
| AURA-248 | Task | Investigate RealSense camera mounting near end-effector | 2 | To Do |
| AURA-249 | Task | Add Zenoh query/export tool for recorded episodes | 2 | To Do |
| AURA-610 | Story | Evaluate LeRobot ACT/DiffusionPolicy for manipulation BC | 3 | To Do |
| AURA-611 | Task | Replace vendor/ar4_ros_driver submodule with upstream | 1 | To Do |
| AURA-711 | Task | Policy Evaluation (sim + sim2real framework) | 3 | To Do |
Data plane (AURA-176) — P2
| Issue | Type | Title | Est | Status |
|---|---|---|---|---|
| AURA-559 | Story | Onboard VisDrone dataset + Rerun video viz notebook | 5 | In Progress |
| AURA-580 | Story | Artifacts table + R2 public storage for Rerun recordings | 3 | In Progress |
| AURA-324 | Story | Evaluate Rerun SDK for lakehouse data visualization | 2 | To Do |
| AURA-326 | Story | Integrate Rerun SDK visualizer into lakehouse demos | 3 | To Do |
| AURA-327 | Story | Demo: agent-controlled Rerun visualizer (NL eval) | 3 | To Do |
| AURA-330 | Task | Sample 25k COCO images via minicoco → HuggingFace | 2 | To Do |
| AURA-570 | Task | Study HF auto-Parquet, Data Studio, dataset viewer | 2 | To Do |
| AURA-589 | Task | Scheduled rclone sync for Paperpile papers every 8h | 1 | To Do |
| AURA-124 | Task | Fix sync to HF Hub | 1 | To Do |
Control plane (AURA-544 orchestrator, AURA-709 algorithm libs) — P2
| Issue | Type | Title | Est | Status |
|---|---|---|---|---|
| AURA-347 | Story | Implement FastAPI job service with Ray actor queue | 5 | To Do |
| AURA-517 | Task | Backend wiring — Postgres, Redis, connect agents to API | 5 | To Do |
| AURA-497 | Task | Aura: Claude Agent SDK orchestrator for physical AI | 5 | To Do |
| AURA-372 | Story | Software architect agent with Compass component sync | 3 | To Do |
| AURA-378 | Story | Jobs page — list jobs and submit form | 5 | To Do |
| AURA-379 | Story | Clusters page — display KubeRay health | 5 | To Do |
| AURA-380 | Story | Experiments page — W&B run browser | 5 | To Do |
| AURA-381 | Story | Live job status updates — polling or WebSocket | 5 | To Do |
| AURA-695 | Task | Define login flow from aegean.ai site to Auraison | 3 | To Do |
| AURA-710 | Story | Perception library — object detection + 6-DoF pose | 5 | To Do |
| AURA-649 | Task | DeepStream-based video ingest service (cross-repo) | 5 | To Do |
Docs & Architecture (AURA-213) — P2/P3
33 open items — lecture/course content (eaia), Compass setup, README/tooling bug fixes, and the Docusaurus → fumadocs migration (AURA-696). Scheduled opportunistically. Highest-value near-term: AURA-205 (control ↔ user plane design doc), AURA-206 / AURA-377 (generative-AI-services ADR — duplicates, reconcile), and Compass setup (AURA-373--AURA-375).
Applications & Research — backlog (not sprint-scheduled yet)
- User Plane: Robotics/Sim (AURA-210, 10 open) — environment-agent + camera-network work (AURA-408--AURA-412), Zenoh bridge bug (AURA-404).
- TurtleBot Maze (AURA-712, 6 open) — RoboMaker worlds in progress (AURA-600, AURA-612), ROS 2 docs MCP server, VSLAM pipeline.
- CUA PDF Reader (AURA-677, 10 open) — course project; OmniDocBench harness, OpenCUA base, Gradio UI.
- DEA / Sidewalk / Tube QC / Counter-UAS (AURA-688, AURA-713, AURA-714, AURA-715) — governing epics, no scoped child work yet.
- Research Book (AURA-716, 26 open) — RL-finetuning and VLM/GRPO evaluations (AURA-618--AURA-627), competitive analyses, and theory threads (score → Fisher → MI). Ongoing; not sprint-bound.
Sprint allocation
Single-assignee capacity (~4 effective days/sprint). Strategy: finish in-progress infra and AR4 sim first (they unblock serving and the manipulation chain), then the AR4 episode/policy chain, then data plane and control-plane wiring.
Sprint 16 (2026-06-30 to 2026-07-11) — Close out in-progress serving + AR4 sim
Sprint 17 (2026-07-14 to 2026-07-25) — AR4 sim landed + Ray cluster
| Issue | Title | Est | Priority |
|---|---|---|---|
| AURA-234 | AR4 Gazebo pick-and-place simulation | 5 | P1 |
| AURA-507 | Single-VM Ray cluster on Proxmox | 3 | P1 |
Sprint 18 (2026-07-28 to 2026-08-08) — AR4 episode chain
| Issue | Title | Est | Priority |
|---|---|---|---|
| AURA-239 | MoveIt pick-and-place demo script | 3 | P1 |
| AURA-238 | Record pick-and-place episodes (lerobot-ros) | 3 | P1 |
Sprint 19 (2026-08-11 to 2026-08-22) — Policy serving + eval
| Issue | Title | Est | Priority |
|---|---|---|---|
| AURA-219 | Policy Server: model-agnostic VLA endpoint | 5 | P1 |
| AURA-711 | Policy Evaluation framework | 3 | P2 |
Sprint 20 (2026-08-25 to 2026-09-05) — Data plane Rerun finish
| Issue | Title | Est | Priority |
|---|---|---|---|
| AURA-559 | VisDrone dataset + Rerun viz notebook | 5 | P2 |
| AURA-580 | Artifacts table + R2 public storage | 3 | P2 |
Sprint 21 (2026-09-08 to 2026-09-19) — Control-plane wiring
| Issue | Title | Est | Priority |
|---|---|---|---|
| AURA-347 | FastAPI job service with Ray actor queue | 5 | P2 |
| AURA-517 | Backend wiring — Postgres, Redis, agents → API | 5 | P2 |
Beyond Sprint 21 (backlog)
- AR4: remaining manipulation chain (AURA-220--AURA-225, AURA-244), GR00T N1 (AURA-224).
- Control plane: frontend pages (AURA-378--AURA-381), orchestrator (AURA-497), algorithm libs (AURA-709).
- Apps: CUA PDF Reader (AURA-677), TurtleBot Maze (AURA-712).
- Research Book (AURA-716) — ongoing, scheduled as slack allows.
Milestones
| Milestone | Target Sprint | Key Deliverable | Epic |
|---|---|---|---|
| M1: Model serving live | Sprint 16 | vLLM via Ray Serve on gpu-node-3 behind CF AI Gateway | AURA-670 |
| M2: AR4 sim ready | Sprint 17 | Gazebo pick-and-place sim running | AURA-216 |
| M3: AR4 dataset + policy | Sprint 19 | Episodes recorded, policy server serving a VLA | AURA-216 |
| M4: Data plane viz | Sprint 20 | VisDrone + Rerun recordings public on R2 | AURA-176 |
| M5: Backend wired | Sprint 21 | Job service + Postgres/Redis + agents on API routes | AURA-544 |
Risks
- Overcommitted sprints: several sprints estimate 8+ days against ~4-day capacity; assumes some tasks beat estimates and parallel work across codebases (infra vs AR4 vs data plane).
- Single assignee: no cross-person parallelism (0.6 availability already factored in).
- AR4 dependency chain: sim → episode logging → policy server → GR00T is sequential; a slip in sim (AURA-234) cascades.
- Hardware/infra: Ray serving and gpu-node-3 availability (AURA-669 Coolify connection) gate M1.
- Docs duplication debt: a few near-duplicate docs tickets survived the orphan triage (e.g. AURA-206 ↔ AURA-377) and should be reconciled.
Notes
- Sprint numbering continues from Sprint 15 (ended 2026-06-27).
- Epic issues are not individually scheduled; they close when all children close.
- Estimates are indicative.
- This plan reflects the post-restructure Jira state: legacy and duplicate work has been archived under AURA-214; research consolidated under AURA-716.