RESEARCH · RESEARCH · #3455
Researchers open-source 'Aspire' agentic robotics framework as AI models speed robot simulation
At the Conference on Robot Learning, researchers including Jim Fan and Ken Goldberg open-sourced Aspire, an agentic framework that helps robots discover and compose new skills. The article's author also reports using OpenAI's GPT-6 Sol alongside tools such as Nvidia Isaac Sim and Newton, Nous Research's Hermes, and an Nvidia DGX Spark to train a virtual Pollen Robotics Microduck to navigate and identify a cat in simulation within hours, illustrating how recent models (and OpenAI's Astra) can generate detailed 3D simulations and automate parts of robot training.
KEY POINTS
- At the Conference on Robot Learning, researchers including Jim Fan and Ken Goldberg open-sourced Aspire, an agentic framework that helps robots discover and compose new skills.
- The article's author also reports using OpenAI's GPT-6 Sol alongside tools such as Nvidia Isaac Sim and Newton, Nous Research's Hermes, and an Nvidia DGX Spark to train a virtual Pollen Robotics Microduck to navigate and identify a cat in simulation within hours, illustrating how recent models (and OpenAI's Astra) can generate detailed 3D simulations and automate parts of robot training.
- Open-sourcing agentic robotics frameworks plus models that can build accurate 3D simulations could greatly shorten the time and expertise needed to develop and transfer complex robot behaviors to the real world.
WHY IT MATTERS
Open-sourcing agentic robotics frameworks plus models that can build accurate 3D simulations could greatly shorten the time and expertise needed to develop and transfer complex robot behaviors to the real world.