1. [Publications](/publications)
2. ACGD: Visual Multitask Policy Learning with Asymmetric Critic Guided Distillation
 
 # ACGD: Visual Multitask Policy Learning with Asymmetric Critic Guided Distillation

  ![](/sites/default/files/styles/wide/public/publications/ACGD_fig.png?itok=mL3I7C_6)

 ACGD introduces a novel approach to visual multitask policy learning by leveraging asymmetric critics to guide the distillation process. Our method trains single-task expert policies and their corresponding critics using privileged state information. These experts are then used to distill a unified multi-task student policy that can generalize across diverse tasks. The student policy employs a VQ-VAE architecture with a transformer-based encoder and decoder, enabling it to predict discrete action tokens from image observations and robot states. We evaluate ACGD on three challenging multi-task domains—MyoDex, BiDex, and OpDex—and demonstrate significant improvements over baseline methods such as BC-RNN+DAgger, ACT, and MT-PPO. ACGD achieves a 10-15% performance boost across various dexterous manipulation benchmarks, showcasing its effectiveness in scaling to high degrees of freedom and complex visuomotor tasks.



 ## Authors



Krishnan Srinivasan (Stanford University)

[Jie Xu](/person/jie-xu)

Henry Ang (Stanford Universidy)

Eric Heiden (NVIDIA)

Dieter Fox (NVIDIA)

Jeannette Bohg (Stanford University)

Animesh Garg (Georgia Tech)

 

 

 ## Publication Date



Sunday, October 19, 2025

 

 ## Published in



[IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025](https://www.iros25.org/)

 

 ## Research Area



[Artificial Intelligence and Machine Learning ](/research-area/machine-learning-artificial-intelligence)

[Robotics](/research-area/robotics)

 

 

 ## External Links



[Project Website](https://critic-guided-distillation.github.io/)