Presented with our partners Austin Robotics and AI

There are many exciting developments in Robotics and AI right now, so let's get together to learn, connect, and create more opportunities for everyone here in Austin.

Talk: Jiaheng Hu: Self-Improving Robots: Towards Practical Reinforcement Learning for General-Purpose Physical AI

Abstract:
The dream of general-purpose robots that can navigate and manipulate our unstructured world is often stymied by a simple reality: we don't have enough robotics data. While Reinforcement Learning (RL) offers a path forward by enabling robots to collect and learn from their own interaction data, existing approaches face significant hurdles in scalability and real-world applicability. In this talk, I will present a suite of research designed to push the frontier of scalable robotic RL. Specifically, I will introduce methods that: (1) leverage inexpensive, "imperfect" simulators to learn structured latent action representations; (2) accelerate learning from sparse feedback by integrating intrinsic motivation and behavior priors; and (3) enable continual RL across diverse, non-stationary tasks without catastrophic forgetting. Together, these contributions represent a significant step toward self-improving robotic systems that progressively expand their capabilities, moving us closer to the realization of general-purpose robotic intelligence.

Bio:
Jiaheng Hu is a 4th-year PhD student at UT-Austin, co-advised by Prof. Peter Stone and Prof. Roberto Martín-Martín. His research interest is in Robot Learning and Reinforcement Learning, with the long-term goal of developing self-improving robots that can learn and adapt autonomously in unstructured environments. Jiaheng’s work has been published at top-tier Robotics and ML venues such as CoRL, NeurIPS, RSS, and ICRA, including a best paper award at ICRA 2024. During his Ph.D., he interned at Google DeepMind and Ai2, and is a recipient of the Two Sigma Ph.D. Fellowship.

SPONSORS
HICAM
A huge shout-out to HICAM for their ongoing support of our community!

Based right here in East Austin, HICAM is a premier 52,000 sq. ft. innovation hub that bridges the gap between digital AI and physical manufacturing. They provide the space, advanced tools, and expertise needed for startups to scale from prototype to production.
We are incredibly grateful for their commitment to fueling the robotics and automation ecosystem in Central Texas.

ANTLER VC

We would like to extend a special thank you to Antler for supporting our community. As the world’s most active inception-stage venture capital firm, Antler partners with exceptional founders from day zero to build and scale category-defining companies. Their commitment to the Austin ecosystem and the AI era helps provide the platform and capital necessary for visionary entrepreneurs to thrive.

Thank you for making our monthly meetups possible!

‍

PromptLab is where ideas get unstuck.

In collaboration with Steven Tran from Soundcheck and Gary Shen from Applied AI Society, PromptLab brings together creative technologists, builders, and thinkers to level up human-AI collaborations.

Bring a project, prototype or concept you’re working on and we’ll explore how to approach it with AI. Together, we’ll break down problems, optimize prompts, and explore new ways to build, design, and think.

This is a space to experiment, iterate, and see what happens when human intuition meets machine capability.

Leave with new tools, new workflows, and a fresh perspective on your project.

​🦞 OpenClaw Architecture — A Phased Learning Journey by Organized AI

​Learn how to build your own personal AI assistant from the ground up — one layer at a time.

​OpenClaw is a local-first AI gateway that connects 20+ messaging channels (Telegram, Slack, Discord, WhatsApp, iMessage, and more) to a single AI agent running entirely on your machine. This isn't a cloud SaaS. It's your own Jarvis — private, configurable, and extensible.

​🗓 Agenda

​First Half — Learn It We'll walk through the full OpenClaw architecture together, phase by phase. Come ready to follow along — no install required for this half.

• ​Welcome — Mental model overview, learning mode selection
• ​Part 1 — Gateway + Channels (the routing layer, adapters, DM pairing)
• ​Part 2 — Agents + Sessions (the 6-step loop, thinking levels, session types)
• ​Part 3 — Skills + Nodes + Security (what it knows, what it can do, how it stays safe)

​~5 min break

​Second Half — Build It Hands-on time. Get OpenClaw running, connect a real channel, and write your first workspace skill.

• ​Setup — Install OpenClaw, clone repo, run pnpm setup, pick your learning mode
• ​Guided Build — Connect a channel, write a workspace skill, run openclaw doctor
• ​Open Floor — Live debugging, Q&A, share what you built

​What You'll Learn

​This session follows the OpenClaw Education curriculum — a 6-phase interactive learning journey:

​🔌 Phase 1 — The Gateway The WebSocket control plane at :18789. Sessions, channel adapters, tool routing, and hot-reload config.

​💬 Phase 2 — Channels 20+ messaging adapters. One internal protocol. The AI never knows which channel sent the message.

​🧠 Phase 3 — Agents + Sessions The Pi agent's 6-step loop: receive → think → tools → stream reply → update session. Thinking levels, session isolation, /compact, and agent-to-agent communication.

​🛠 Phase 4 — Skills + Nodes Skills = what your agent knows. Nodes = what your agent can do. macOS, iOS, and Android companion apps exposing camera, shell, location, and more via node.invoke.

​🔐 Phase 5 — Tailscale + Security Six security layers. Default-safe, opt-in to exposure. Tailscale Serve, Funnel with enforced auth, and Docker sandboxing for group sessions.

​🚀 Phase 6 — Deployment Deploy your own doc site to Cloudflare Pages, run a remote Linux gateway, or use Docker Compose or Nix. Leave with a running assistant.

​Three Learning Modes

​📖 Docs Only — No install needed. Diagrams and concept explainers. ⭐ Diagrams + Docs — Mermaid diagrams and annotated examples. Recommended starting point. 💻 Full Hands-On — Live Gateway + channel exercises. OpenClaw required.

​Prerequisites Node.js 22.16+, pnpm 8+, Git. OpenClaw optional depending on your learning mode.

​git clone https://github.com/Organized-AI/openclaw-educationcd openclaw-education && pnpm install && pnpm setup

​Follow the full guide → openclaw-education.pages.dev

​Your Presenter

​Jordan Hill is the founder of Organized AI and Blue Highlighted Text, an AI automation and consulting agency based in Austin, TX. He builds agentic infrastructure for real businesses — from conversion tracking pipelines to self-hosted AI assistants — and created the OpenClaw Education curriculum to make personal AI architecture accessible to anyone willing to get their hands dirty.

​Organized AI · The Lobster Way 🦞

‍

The Lovable Founder Series is coming to Austin, hosted by Steven Tran!

The Founder Series is bringing together aspiring and early-stage founders across cities worldwide - people who want to turn ideas into real businesses using Lovable.

More info coming soon!