VLAICLR 20252025
Pi-0.5: A Foundation Policy Model for Generalist Robots
Physical IntelligencePhysical Intelligence
摘要
We present Pi-0.5, a foundation policy model that enables generalist robot behavior across diverse embodiments and tasks. Pi-0.5 is pre-trained on large-scale robot data and fine-tuned for specific applications, achieving robust performance on manipulation, navigation, and mobile manipulation tasks.
foundation policygeneralist robotdiverse embodimentpre-trainingrobust performance
技术细节
数据集
Open X-EmbodimentPhysical Intelligence Data
模型骨架
Pi-0.5 Foundation Model (VLM + Flow Matching)
编码器
SigLIP + PaliGemma + Proprioceptive Encoder
解码器
Flow Matching Action Decoder (continuous actions)