Generalist AI releases GEN-1.5 embodied foundation model learning physical tasks in seconds
Generalist AI released GEN-1.5 on Aug 23, an embodied foundation model trained on 8 months of physical interaction, achieving 59% one-shot success.
On August 23, Generalist AI released GEN-1.5, an embodied foundation model. The model is based on eight months of continuous pretraining on large-scale physical interaction data and can learn physical tasks in seconds without explicit task-specific training.
GEN-1.5 achieves 59% success rate with one-shot in-context learning from 3-12 second demonstrations, and 83% success after 10 gradient steps on 5 minutes of data. The model shows particular strength in robot manipulation, operational workflows, and multimodal perception tasks.
Generalist AI was founded by well-known embodied intelligence researchers including Pieter Abbeel and Sergey Levine. GEN-1.5 reaches the frontier of open-source models on benchmarks such as Physical Intelligence.