Open-Source Arms, Reward Hacks, and the Autodidact Problem

Research

Active Autodidactic Learning for Home Robots

Home robots that actively seek out and practice tasks they struggle with instead of waiting for human data. This is how you get robots that actually improve in the field without constant babysitting.

Rapid Sim-to-Real Adapters for Quadrotor Control

A new method that lets you adapt a sim-trained controller to a real quadrotor in minutes, not hours. This directly attacks the sim-to-real bottleneck that kills most real-world robotics projects.

LLM-Designed Rewards for Complex Locomotion

Using a large language model to design reward functions for robots that walk. This is a huge leverage point: automating the painful, manual part of reinforcement learning.

Tools

Open-Source 6-DOF Manipulation Platform

A fully open-source, low-cost arm for serious manipulation research. The barrier to entry for building and testing manipulation algorithms just dropped significantly.

Isaac Sim 5.0: New Thermal Calibration Tools

Nvidia's latest Isaac Sim update adds tooling for thermal calibration, a niche but critical detail for reliable real-world sensor fusion. The sim-to-real gap keeps shrinking.