Rohan Siva

I am pursuing a B.S. in Electrical and Computer Engineering Honors with a Minor in Stats & Data Science at the University of Texas at Austin, graduating in May 2027.

My research focuses on robotics safety, reinforcement learning, and uncertainty quantification, applied to world models, vision-language-action models, diffusion models, and foundation model post-training and inference. I build embodied, neurosymbolic systems that perceive, reason, and act safely in complex environments.

Email Github Scholar LinkedIn

News

Sep 2026 My first-author paper "What Objects Enable, Not What They Are: Functional Latent Spaces for Affordance Reasoning" was accepted to the CoRL 2026 main conference! 🏆
Aug 2026 Wrapped up my research internship at Decagon AI, publishing two blog posts: Teaching Flow-Matching Text-to-Speech Models with RL and Scaling Real-Time Text-to-Speech Inference.
May 2026 Our paper "VLN-Zero: Rapid Exploration and Cache-Enabled Neurosymbolic Vision-Language Planning for Zero-Shot Transfer in Robot Navigation" was selected as a Spotlight at NEUS 2026. 🏆
Dec 2025 Our paper "UNCAP: Uncertainty-Guided Planning Using Natural Language Communication for Cooperative Autonomous Vehicles" was accepted to AAMAS 2026 as an Oral + Best Paper Finalist. 🏆
Aug 2025 Wrapped up my internship at Cisco with the HyperAI Lab team, where I worked on kRAIG, a natural language-driven agent for automated DataOps pipeline generation.
Feb 2025 Our paper "Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework" was accepted to MLSys 2025 as an Oral. 🏆

Publications

What Objects Enable, Not What They Are: Functional Latent Spaces for Affordance Reasoning

CoRL 2026 🏆
Siva, R., Bhatt, N. P., Yang, Y., Lee, S., Gadde, N., Ellis, C., Velasquez, A., Wang, Z., Topcu, U.

A4D maps visual observations into a shared functional latent space structured around affordances (e.g., "movable"), enabling robot planning based on task-relevant object functionalities rather than appearance alone. We achieve 94% inference accuracy on existing affordances — over 15% above state-of-the-art — while enabling 100x faster inference.

Project Page arXiv Code

Scaling Real-Time Text-to-Speech Inference

Industry Blog
Siva, R., Asgari, C.

Rebuilding Decagon's streaming TTS pipeline on M* with stage-aware scheduling, packed prefill, paged KV caching, and CUDA Graphs — achieving 3x lower TTFA (27 ms) and 10x higher throughput at concurrency 8 for real-time voice model inference.

Blog Post X

Teaching Flow-Matching Text-to-Speech Models with RL

Industry Blog
Siva, R., Asgari, C.

Flow-matching TTS architectures lack standard open-source RL post-training support. We design a specialized RL framework (GRPO/PPO) for continuous flow-matching models, eliminating acoustic artifacts and tightening sample distribution variance.

Blog Post X

Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework

MLSys, 2025 (Oral) 🏆
Bhatt, N. P., Yang, Y., Siva, R., Milan, D., Topcu, U., Wang, Z.

A novel framework for uncertainty-aware multimodal planning using conformal prediction for perception uncertainty and FMDP to quantify decision uncertainty, with formal verification guarantees. Building on this, we implement active sensing and automated refinement via SFT to meet task specifications, reducing variability by 40% and improving task success by 5%.

Project Page arXiv Code OpenReview

UNCAP: Uncertainty-Guided Planning Using Natural Language Communication for Cooperative Autonomous Vehicles

AAMAS, 2026 (Oral + Best Paper Finalist) 🏆
Bhatt, N. P., Li, P., Gupta, K., Siva, R., Milan, D., Hogue, A. T., Chinchali, S. P., Fridovich-Keil, D., Wang, Z., Topcu, U.

A framework for uncertainty-guided planning in cooperative autonomous vehicles using natural language communication. Leverages uncertainty quantification to improve coordination and decision-making in multi-agent autonomous driving scenarios.

Project Page arXiv Code OpenReview

VLN-Zero: Rapid Exploration and Cache-Enabled Neurosymbolic Vision-Language Planning for Zero-Shot Transfer in Robot Navigation

NEUS, 2026 (Spotlight) 🏆
Bhatt, N. P., Yang, Y., Siva, R., Samineni, P., Milan, D., Wang, Z., Topcu, U.

A neurosymbolic approach combining vision-language models with cache-enabled planning for zero-shot robot navigation. Enables rapid exploration and transfer learning in novel environments without task-specific training.

Project Page arXiv Code

kRAIG: A Natural Language-Driven Agent for Automated DataOps Pipeline Generation

Under Submission
Siva, R., Cheung, K., Li, L., Sundaram, G.

An AI agent that translates natural language specifications into production-ready Kubeflow Pipelines. Introduces ReQuesAct to clarify user intent prior to pipeline synthesis, with retrieval-augmented tool generation and LLM-based validation. Achieves a 3x improvement in extraction/loading success and 25% higher transformation accuracy over state-of-the-art agentic baselines.

arXiv

RepV: Safety-Separable Latent Spaces for Scalable Neurosymbolic Plan Verification

Under Submission
Yang, Y., Bhatt, N. P., Samineni, P., Siva, R., Wang, Z., Topcu, U.

A novel framework for safety-verifiable reinforcement learning by learning safety-separable latent spaces that enable efficient neurosymbolic plan verification. Combines the representational power of deep learning with the formal guarantees of symbolic methods for scalable safety verification in complex environments.

Project Page arXiv Code

Co-GLANCE: Uncertainty-Aware Active Perception for Heterogeneous Robot Teaming

Under Submission
Podolinsky, M. P., Bhatt, N. P., Samineni, P., Siva, R., Ellis, C., Topcu, U.

A real-time onboard perception and decision-making system for uncertainty resolution in heterogeneous robot teams. Distills semantic reasoning from vision-language models into an end-to-end model for occlusion segmentation and robot allocation, combining selective abstention and conformal prediction for calibrated uncertainty guarantees.

Project Page arXiv Code