The startup Figure has unveiled Helix 2.5, a new neural network model designed to enable humanoid robots to operate in unfamiliar home environments. The system allows machines to perform complex household chores without prior training or specific adaptation to a new space.
During testing, the robot successfully navigated 30 previously unseen homes. The machine demonstrated the ability to make beds, fold towels, and organize living rooms while interacting with objects it had never encountered before. The system is designed to identify and correct its own errors in real-time without human intervention.
Figure emphasized that the neural network underwent no fine-tuning for the test environments. The company did not collect data from these specific houses prior to the trials. This zero-shot capability aims to eliminate the need for custom programming for every individual household, allowing robots to adapt to new surroundings autonomously.
