cv

MS in Computer Science

University of California San Diego

• Specialization: Artificial Intelligence and Machine Learning
• GPA: 4.0/4.0

B.Tech in Aerospace Engineering, Minor in Machine Learning

Indian Institute of Technology Kanpur

• CPI: 8.9/10

Cross-Embodiment Laws for Robot Manipulation

Research Intern with Professor Hao Su

• Key Contribution: Developing a pipeline for scaling Reinforcement Learning (RL) across diverse robot embodiments using a single policy. Designed a novel architecture improving cross-embodiment generalization by 10%, currently working on developing novel representations for robots.

Controllable Image Generation via Joint Distribution Learning

Research Intern with Professor Zhuowen Tu

• Key Contribution: Developing a novel curriculum learning strategy to enforce layout-based controllability in diffusion models enabling development of controlled generative models.

Generalist Vision-Language-Action (VLA) Models for Robotics

Research Intern with Professor Hao Su

• Key Contribution: Introduced a dual-encoder VLA architecture with a language-aligned action tokenizer that preserves pretrained VLM features, boosting generalization and increasing task success by 40% over SOTA baselines.

Benchmarking and Enhancing Corrective Sequential Planning in VLMs

Research Intern with Professor Yogesh Rawat and Vibhav Vineet

• Key Contribution: Created the CosPlan benchmark for corrective sequential planning and proposed the Scene Graph Incremental (SGI) update method, a novel training-free technique that improves error correction by 5.2%.

Improving Temporal Reasoning by Mitigating Hallucinations

Research Intern with Professor Yogesh Rawat and Vibhav Vineet

• Key Contribution: Proposed the Dream of Thoughts (DoT) prompting technique, a novel multi-step reasoning approach that navigates model hallucinations to significantly improve the understanding of activity transitions.

Robust Video Action Recognition under Occlusion

Research Intern with Professor Yogesh Rawat and Vibhav Vineet

• Key Contribution: Designed the CTx-Net compositional architecture for robust video understanding and introduced three benchmark datasets to systematically study and evaluate model performance under occlusion.

CoDiffusion: Harmonizing Layouts and Image Generation through Joint Diffusion

• S. Grover, S. Aggarwal, D. Srivastav, Z. Tu

Enhancing Generalization in VLA Models by Preserving Pretrained Representations

• S. Grover, A. Gopalkrishnan, Bo Ai, H. I. Christensen, H. Su, Xuanlin Li

Navigating Hallucinations for Reasoning of Unintentional Activities

• S. Grover, V. Vineet, Y. S. Rawat

CosPlan: A Benchmark for Corrective Sequential Planning

• S. Grover, V. Vineet, Y. S. Rawat

Revealing the Unseen: Benchmarking Video Action Recognition under Occlusion

• S. Grover, V. Vineet, Y. S. Rawat