PhD Candidate · MIT MechE & AeroAstro · REALM Lab

Reliable AI through
targeted evaluation

I develop principled methods to discover consequential failures in autonomous systems, evaluate behavior against human values, and guide system improvement and perform repairs.

Seeking full-time research and industry roles starting May/June 2027.

Publications CV Email
Anjali Parashar

Research Framework

Discover consequential failures. Learn from limited evidence. Improve system behavior.

01 — Discover Adaptive testing for
efficient failure discovery Find diverse failures under limited testing budgets
02 — Evaluate Human-informed
evaluation Test behavior against contextual goals and stakeholder values
Representative work — click to explore
SEED-SET (ICLR '26) Contextual failures (CoRL '25)
03 — Improve Learning and
optimization Use evidence to guide better decisions and system repair
Shared mathematical foundations
Bayesian inference Experimental design Sampling Optimization
Applications: robotics · autonomous driving · AI decision-making

Updates

News

Oct 2026
Paper

"What Would Removing This Corruption Change? Failure Attribution for Vision-Language-Action Models," led by Carson Sobolewski, accepted at the SPAIS Workshop at CoRL 2026!

Oct 2026
Award

Received Top Reviewer Award at NeurIPS 2026!

Oct 2026
Paper

Paper on adaptive test case discovery for data-efficient LLM evaluation accepted at NeurIPS 2026.

Oct 2026
Award

Received CoRL 2026 Travel Grant for work on coverage-aware failure discovery with paired systems.

Oct 2026
Paper

Coverage Aware Active Evaluation for Failure Discovery with Paired Systems accepted at CoRL 2026. arXiv

Aug 2026
Paper

"Generative Framework to Augment Weather-Driven Disruption Scenarios for Air Traffic Management Decision Support" accepted at AIAA SciTech 2027.

Jun 2026
Internship

Started research internship at NVIDIA ASPIRE Group (Autonomous Systems and Physical AI Research), Santa Clara, working on data-mixing algorithms for VLM fine-tuning in safety evaluation.

May 2026
Paper

Scenario Generation for Testing of Autonomous Driving Systems accepted at NeuS 2026. arXiv

Jan 2026
Paper

SEED-SET accepted at ICLR 2026 — scalable evolving experimental design for system-level ethical testing. arXiv

Jan 2026
Feature

Work on SEED-SET featured in MIT News — experimental design for autonomous system ethical testing.

Oct 2025
Paper

Failure Prediction from Limited Hardware Demonstrations presented at Allerton Conference 2025 by Prof. Chuchu Fan.

Sep 2025
Paper

Cost-aware Discovery of Contextual Failures accepted at CoRL 2025. PDF

May 2025
Paper

RADIUM published in IEEE Transactions on Robotics 2025 — predicting and repairing end-to-end robot failures using gradient-accelerated sampling.

Apr 2025
Talk

Presented Learning-based Bayesian Inference for Testing of Autonomous Systems at ICRA 2025.

Dec 2024
Talk

Presented at NeurIPS BDU Workshop 2024 on failure prediction from limited hardware demonstrations.

Jan 2024
Paper

Learning-based Bayesian Inference for Testing published in IEEE RA-L 2024.

Jun 2023
Update

Started PhD at MIT MechE & AeroAstro, joining the REALM Lab with Prof. Chuchu Fan.

Selected work

Publications

2026–2027
  1. NeurIPS 2026 A. Parashar, C. Sobolewski, Y. Li, F. Chen, C. Fan. "Adaptive Test Case Discovery for LLM-Assisted Decision Making in High-Stakes Domains."
  2. CoRL 2026 A. Parashar, R. Luo, A. Sharma, S. Veer, E. Schmerling, C. Sobolewski, M. Yu, C. Fan, M. Pavone. "Coverage Aware Active Evaluation for Failure Discovery with Paired Systems." [arXiv]
  3. NeuS 2026 A. Parashar, C. Fan. "Scenario Generation for Testing of Autonomous Driving Systems Using Real-World Failure Records." [arXiv]
  4. ICLR 2026 A. Parashar, Y. Li, E. Y. Yu, F. Chen, J. Neidhoefer, D. Upadhyay, C. Fan. "SEED-SET: Scalable Evolving Experimental Design for System-level Ethical Testing." [arXiv] [Website]
  5. TMLR 2026 Y. Li, A. Parashar, C. Fan. "Curiosity is Knowledge: Self-Consistent Learning and No-Regret Optimization with Active Inference." Under review. [arXiv]
  6. AIAA SciTech 2027 M. Borrero, A. Parashar, J. Zhang, J.J. Jones, C. Fan, M. Li. "A Generative Framework to Augment Sparse Weather-Driven Disruption Scenarios for Air Traffic Management Decision Support."
2024–2025
  1. CoRL 2025 A. Parashar, J. Zhang, Y. Li, C. Fan. "Cost-aware Discovery of Contextual Failures using Bayesian Active Learning." [PDF]
  2. Allerton 2025 · NeurIPS BDU Workshop 2024 A. Parashar, J. Zhang, K. Garg, C. Fan. "Failure Prediction from Limited Hardware Demonstrations." [arXiv]
  3. IEEE T-RO 2025 C. Dawson, A. Parashar, C. Fan. "RADIUM: Predicting and Repairing End-to-End Robot Failures using Gradient-Accelerated Sampling." [PDF]
  4. TMLR 2025 T.E. Gibson, S. Acharya, A. Parashar, J.E. Gaudio, A.M. Annaswamy. "On the stability of gradient descent with second order dynamics for time-varying cost functions." ICLR 2026 J2C certification. [PDF]
  5. IEEE RA-L 2024 A. Parashar, J. Yin, C. Dawson, P. Tsiotras, C. Fan. "Learning-based Bayesian Inference for Testing of Autonomous Systems." Presented at ICRA 2025. [RA-L] [Website]
  6. Field Robotics 2024 A. Saba et al. (incl. A. Parashar). "Fast and modular autonomy software for autonomous racing vehicles."
2020–2023
  1. ACC 2023 A. Parashar, P. Srivastava, A.M. Annaswamy. "Accelerated Algorithms for a Class of Optimization Problems with Equality and Box Constraints."
  2. CDC 2022 A. Parashar, P. Srivastava, A.M. Annaswamy, B. Dey, A. Chakraborty. "Accelerated Algorithms for a Class of Optimization Problems with Constraints."
  3. Ocean Engineering 2020 J. Kadiyam, A. Parashar, S. Mohan, D. Deshmukh. "Actuator fault-tolerant control study of an underwater robot with four rotatable thrusters."

Full list on Google Scholar.

About

Anjali Parashar
she/her · PhD Candidate
MIT MechE & AeroAstro, REALM Lab
Advisor: Prof. Chuchu Fan

My research develops principled methods for testing and evaluating autonomous systems. This involves efficiently finding diverse scenarios where they fail and quantifying behavior relative to human values through scenario generation. I leverage several Bayesian inference tools, such as Bayesian Experiment Design for user-in-the-loop design of scenarios and MCMC sampling combined with optimization for targeted exploration.

Previously, I finished my Master's thesis on optimization and control with Dr. Anuradha Annaswamy (2021–23), and Bachelor's thesis on fault-tolerant control of underwater vehicles under Dr. Santhakumar Mohan (2020).

CV Google Scholar anjalip@mit.edu
Add paper figure or video here