Anca Dragan
Associate Professor
Works on AI alignment and human-robot interaction, learning from human feedback and modeling human decision-making.
Anca Dragan's lab at UC Berkeley works on human-robot interaction & assistive robots, reinforcement learning and trustworthy & safe ML. 37 papers since 2023, 35 at top venues, mostly ICLR and ICML. 8 PhD students. 42 alumni, most often to Google DeepMind and Freedom Robotics. Applications close .
At a glance
- papers, 2023–2026
- 10/ year
- at T1 venues
- 95%
- PhD students
- 8
- MS students
- 0
- alumni listed
- 42
Papers by yearfrom DBLP, 2023–2026
- 0 papers2026 · T1 0 · collecting
- 8 papers2025 · T1 8
- 15 papers2024 · T1 15
- 14 papers2023 · T1 12
T1 venuesOtherVenues: ICLR, ICML, NeurIPS, CoRL, HRI
Research focuspaper titles that use each topic's words
- Reinforcement learning22 papers
- Human-robot interaction & assistive robots10 papers
- human8 papers
- human feedback2 papers
- Trustworthy & safe ML1 paper
Paper list
From DBLP.
20258 papers
- ICML (T1)Adversaries Can Misuse Combinations of Safe Models (opens in a new tab)
- ICML (T1)AssistanceZero: Scalably Solving Assistance Games (opens in a new tab)
- ICLR (T1)Context Steering: Controllable Personalization at Inference Time (opens in a new tab)
- ICLR (T1)Correlated Proxies: A New Definition and Improved Mitigation for Reward Hacking (opens in a new tab)
- ICLR (T1)On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback (opens in a new tab)
- NeurIPS (T1)Planning without Search: Refining Frontier LLMs with Offline Goal-Conditioned RL (opens in a new tab)
- ICLR (T1)Q-SFT: Q-Learning for Language Models via Supervised Fine-Tuning (opens in a new tab)
- NeurIPS (T1)Temporal Representation Alignment: Successor Features Enable Emergent Compositionality in Robot Instruction Following (opens in a new tab)
202415 papers
- ICRA (T1)A Generalized Acquisition Function for Preference-based Reward Learning (opens in a new tab)
- ICML (T1)AI Alignment with Changing and Influenceable Reward Functions (opens in a new tab)
- HRI (T1)Aligning Human and Robot Representations (opens in a new tab)
- ICLR (T1)Confronting Reward Model Overoptimization with Constrained RLHF (opens in a new tab)
- ICML (T1)Coprocessor Actor Critic: A Model-Based Reinforcement Learning Approach For Adaptive Brain Stimulation (opens in a new tab)
- AAMAS (T1)Defining Deception in Decision Making (opens in a new tab)
- AAAI (T1)Learning Optimal Advantage from Preferences and Mistaking It for Reward (opens in a new tab)
- ICML (T1)Learning Temporal Distances: Contrastive Successor Features Can Provide a Metric Structure for Decision-Making (opens in a new tab)
- NeurIPS (T1)Learning to Assist Humans without Inferring Rewards (opens in a new tab)
- ICML (T1)Learning to Model the World With Language (opens in a new tab)
- ICLR (T1)Offline RL with Observation Histories: Analyzing and Improving Sample Complexity (opens in a new tab)
- ICLR (T1)The Effective Horizon Explains Deep RL Performance in Stochastic Environments (opens in a new tab)
202314 papers
- ICML (T1)Automatically Auditing Large Language Models via Discrete Optimization (opens in a new tab)
- Trans. Mach. Learn. Res.Benchmarks and Algorithms for Offline Preference-Based Reward Learning (opens in a new tab)
- IROS (T1)Bootstrapping Adaptive Human-Machine Interfaces with Offline Reinforcement Learning (opens in a new tab)
- NeurIPS (T1)Bridging RL Theory and Practice with the Effective Horizon (opens in a new tab)
- ICLR (T1)Causal Confusion and Reward Misidentification in Preference-Based Reward Learning (opens in a new tab)
- ICML (T1)Contextual Reliability: When Different Features Matter in Different Contexts (opens in a new tab)
- CoRL (T1)Goal Representations for Instruction Following: A Semi-Supervised Language Interface to Control (opens in a new tab)
- NeurIPS (T1)Learning to Influence Human Behavior with Offline Reinforcement Learning (opens in a new tab)
- ICLR (T1)On the Sensitivity of Reward Inference to Misspecified Human Models (opens in a new tab)
- Trans. Mach. Learn. Res.Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback (opens in a new tab)
- CoRL (T1)Quantifying Assistive Robustness Via the Natural-Adversarial Frontier (opens in a new tab)
- HRI (T1)SIRL: Similarity-based Implicit Representation Learning (opens in a new tab)
Where alumni went42 listed, 29% to industry
Industry 4Academia 10
Most often to
- Google DeepMind
- Freedom Robotics
- the Partnership for AI
Also to
- Carnegie Mellon University
- MIT
- Stanford University
Apply to UC BerkeleyFall 2027
UC Berkeley EECS Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- not required
- English
- TOEFL/IELTS (non-English-country degrees)
- Fee
- See program page
Confirmed on the program page.
Similar labsnearest by research summary
- CNTsinghua
Hui Qiao
- Medical imaging & healthcare AI
- DETUM
Daniel Rueckert
- Medical imaging & healthcare AI
- USMichigan
Liyue Shen
- Medical imaging & healthcare AI
- GBOxford
Ana Namburete
- Neuroscience & brain-computer interfaces
- HKCUHK
Pheng Ann Heng
- Medical imaging & healthcare AI
- GBOxford
Irina Voiculescu
- Medical imaging & healthcare AI
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- DBLP
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
UC Berkeleycloses