Icml26

Princeton RL @ ICML 2026! I’m excited to share progress we’ve made in RL algorithms:

• On the Role of Computation in Reinforcement Learning, led by Raj Ghugare, with Michał Bortkiewicz and Alicja Ziarko. (spotlight)
• Learning to Perceive the World Through Control: Empowerment-Based Representation Learning, led by Mahsa Bastankhah and Sophie Broderick.
• Training LLM Agents to Empower Humans, led by Evan Ellis, with Vivek Myers, Jens Tuyls, Sergey Levine, and Anca Dragan.
• Consistent Zero-Shot Imitation with Contrastive Goal Inference, led by Kathryn Wantlin, with Chongyi Zheng.

We’re also presenting preliminary work at the workshops:

• Towards Adapting Contrastive RL to the Offline Setting. Led by Catherine Ji, with Grace Tan. (Decision-Making from Offline Datasets to Online Adaptation)
• Can We Really Learn One Representation to Optimize All Rewards?. Led by Chongyi Zheng, with Royina Karegoudra Jayanth. (Decision-Making from Offline Datasets to Online Adaptation (spotlight); RL from World Feedback)
• Is Temporal-Difference Learning the Only Path to Stitching in RL?. Led by Michał Bortkiewicz, with Władysław Pałucki and Mateusz Ostaszewski. (Decision-Making from Offline Datasets to Online Adaptation)