Berkeley's CLIFT enables closed-loop fine-tuning for closed-source humanoid models
keerthanpg · x · 2026-08-20
UC Berkeley, in collaboration with Google DeepMind and NVIDIA, published "CLIFT," a method for Closed-Loop Iterative Fine-Tuning that does not require access to internal model parameters (gradients, losses).
Problem & Motivation:
- Closed-weight robot foundation models (e.g., Gemini Robotics On-Device) are generalists but lack task mastery for agile, contact-rich humanoid tasks due to covariate shift.
- Mastery requires closed-loop learning, but closed-source models only expose a managed SFT API, making standard RL (like PPO) impossible.
Core Method:
- The key insight is that reinforcement signals can be encoded directly into supervised training data.
- By using only the SFT API as a black-box operator mapping observation–instruction–action chunks to a tuned policy, the system achieves RL-style policy improvement without modifying the model or training procedure.
Results:
- This method successfully transforms a small, on-device model based on open-source Gemma into a humanoid specialist, outperforming some of the largest action models on manipulation tasks.
More from Embodied
- Google's RT-2 packed 55B parameters, 1000x RT-1's 55M just 7 months later — binarybits · 2026-08-20
- Perceptron Inc releases SotA VLA model, unifying perception and control — code_star · 2026-08-20
- Turning a $10 toy into an LLM-powered robot: full build guide — calabi_and_yau · 2026-08-20
- Remote delivery breakthrough: controlling a robot from 4000 miles away — StewartalsopIII · 2026-08-20
- GeneralistAI demos GEN-1.5 robot capable of one-shot learning — teortaxesTex · 2026-08-20
- Wired: Robot Improvises Using Banana as Tool in Live Demo — nordicinst · 2026-08-20