Rohan Siva (@_rsiva) profile picture

research @decagonai ml & robotics research @utaustin

Joined August 2025
Photo by Rohan Siva (@_rsiva)
Show a robot a cart it's never seen. It doesn't need to know it's a "cart," just that it's pushable, rolls & holds stuff. Happy to announce my first-author paper on functional latent spaces got into #CoRL2026: "What Objects Enable, Not What They Are" arxiv.org/abs/2606.05533 🧵
3
4
1
20
1,236
Instead of appearance, A4D maps images and affordances into a shared functional latent space. Each affordance gets an axis (e.g. movable ↔ fixed), and we measure how closely an image projects onto that axis, making inference 100x faster than VLM baselines, at ~22ms. (2/4)
1
1
5
243
multimodal inference feels pretty underexplored, especially when you move beyond standard transformer architectures. fun to share what we learned when scaling our text to speech model to 10x throughput
Fast text serving ≠ fast speech. Here's how we got our time to first audio under 30ms with nearly 10x more audio throughput.
3
1
20
1,671
tts naturalness is hard. sft has its limits, but rl is tricky for flow-matching models since they lack the log-probs dpo and grpo rely on. excited to share work on adapting both methods for TTS with @cyrusasg at @DecagonAI!
Modern TTS models can sound great — and still fail badly on pacing, pauses, and prosody. We adapted DPO + GRPO to flow-matching models to tackle the tail end of TTS behavior:
2
21
1,188