Skip to content

Commit 98a56bc

Browse files
committed
docs: publish project status report
1 parent f8c9548 commit 98a56bc

1 file changed

Lines changed: 28 additions & 0 deletions

File tree

‎README.md‎

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,34 @@
44
55
IgnitionRL is not “just another RL library”. The product goal is to let developers, designers, researchers and AI builders iterate on environments, observations, actions, rewards and scenarios without rewriting RL algorithms.
66

7+
## Project status
8+
9+
Status as of 2026-05-29: **paused / public research artifact**.
10+
11+
This repository is being made public after validating the initial startup wedge.
12+
The technical prototype is useful, but the first market thesis was not strong
13+
enough to keep pursuing as a company.
14+
15+
Market validation verdict:
16+
17+
- Robotics and autonomous driving remain interesting long-term markets, but not
18+
credible first wedges for this stack.
19+
- Serious robotics and AD teams need high-fidelity physics, sensor simulation,
20+
ROS/Isaac/MuJoCo/CARLA workflows and sim-to-real credibility. A Three.js/R3F
21+
stack cannot win that market first.
22+
- The viable near-term wedges are robotics education, hobbyist prototyping,
23+
indie simulation/game agents and browser-based RL demos.
24+
- Those wedges are real, but they are not the founder's desired first market, so
25+
the project is paused rather than forcing a compromise.
26+
27+
Reusable assets:
28+
29+
- TypeScript environment contract, runners and policy interfaces.
30+
- Named reward terms, reward debugging concepts and replay traces.
31+
- Local run artifacts: metrics, checkpoints, reports and Studio-ready JSON.
32+
- Tauri/React Studio workflow for inspecting runs and episodes.
33+
- Rust JSONL engine, IPC protocol and Burn/Candle backend research.
34+
735
## Product thesis
836

937
Modern AI can generate React Three Fiber environments very quickly. What is still painful is turning those environments into reliable training loops: observations, rewards, agent decisions, rollout collection, training metrics, checkpoints, inference, curriculum and experiment comparison.

0 commit comments

Comments
 (0)