What is Awesome Astra Embodied AI?

Awesome Astra Embodied AI is a curated list — a set of links and notes — that collects demonstrations, experiments, and benchmark results of a specific AI model called GPT-6 Astra applied to robotics and embodied AI. “Embodied AI” means AI that can sense, think, and act in the physical world, usually through robots.

The list is maintained by a developer (zjwzcx) on GitHub, and anyone can suggest additions or corrections. It is formatted as a structured list of “cases”. Each case links to an external source (a social media post, a GitHub repo, a paper, a benchmark website) and gives a short description, the publication date, and the credits for who created the case.

The cases are grouped into sections by kind of workflow:

• Public Benchmark & Evaluation Reports – independent reports that test GPT-6 Astra on standard robotics tasks and publish scores.
• Zero-shot Control – cases where GPT-6 Astra directly controls a robot without any prior training on that specific task. This section splits into Deploy in Simulation (tasks done in virtual environments) and Deploy in Real World (real robots).
• Agentic Policy Calls – cases where GPT-6 Astra acts more like a planner or “agent” and calls other robot policies.
• Real-to-sim Replay / Data Rollout – using real-world data to recreate simulations or generate training data.
• Astra Builds RL Training Environments and Training – examples of GPT-6 Astra building reinforcement learning environments or training robots.

Each case has a short headline, the source/credit (who did it and where it was posted), the publication date, and a one-sentence summary of what was achieved. The list also includes screenshots of the demonstrations.

In addition, the Benchmark & Evaluation Reports section lists external evaluations with detailed metrics. For example, one report shows GPT-6 Astra completing 19 out of 20 bowl-manipulation tasks, another reports a 92.63% success on a LIBERO benchmark, and a third shows the model ranking 2nd among 9 agents on a video-to-simulation reconstruction benchmark. All these numbers come directly from the linked sources, and the README reminds readers to interpret them within each project's own setup.

Overall, this project is a compilation of community work showcasing what GPT-6 Astra can do in robotics — both in simulation and on real hardware — and how its performance compares across different tasks and benchmarks. It serves as a quick reference for researchers, engineers, and enthusiasts to see the current state of the model's capabilities in embodied AI.