Srikrishna Iyer

I am currently an Assistant Principal AI Engineer at the Next-gen Edge AI & Robotics (NEAR) lab under the Group Technology Office at ST Engineering. I work on research translation, bridging academic research to real-world applications across ST Engineering's business units. My applied research focuses on robotic manipulation, locomotion and navigation in quadrupeds and AMRs.

I got my Masters in Control & Automation at Nanyang Technological University under the supervision of Muhammad Faeyz Karim, and a Bachelors in Electronics & Communications from VIT University.

Happy to chat more about anything related to AI or robotics!

Srikrishna Iyer standing on a flower-lined street

News

  • RL²-VLA, on test-time scaling for vision–language–action models, is under review at IEEE RA-L.
  • Exhibiting Robust Physical AI at the Singapore Airshow, Changi Exhibition Centre (3–8 Feb).
  • Paper on LLMs for forecasting traffic-incident impact presented at IEEE ITSC 2025.
  • When Babies Teach Babies published in the BabyLM Challenge proceedings, CoNLL @ EMNLP 2024.

Publications

2026

RL²-VLA title card beside a robot arm manipulating a red toy car on a workbench

RL²-VLA: Adaptive RL Latent Compositional Steering with Test-Time Scaling for Vision-Language-Action Models

arXiv preprint arXiv:2607.26991 · IEEE Robotics and Automation Letters (RA-L), 2026

Under review

2025

Congestion map of the Los Angeles road network above the question: can state-of-the-art LLMs forecast traffic incident impact without training on incident-specific data?

Application and Evaluation of Large Language Models for Forecasting the Impact of Traffic Incidents

IEEE International Conference on Intelligent Transportation Systems (ITSC), 2025

2024

Method diagram: Bayesian optimisation initialises diverse student models, then a bi-level loop weights their mutual learning against a teacher

When Babies Teach Babies: Can Student Knowledge Sharing Outperform Teacher-Guided Distillation on Small Datasets?

The 2nd BabyLM Challenge at CoNLL, Miami FL, 2024 · pp. 197–211

2023

GAT-GAN architecture: the training objective beside encoder, decoder and discriminator stacks, each built from 1D convolutions with spatial and temporal attention

GAT-GAN: A Graph-Attention-based Time-Series Generative Adversarial Network

arXiv preprint arXiv:2306.01999, 2023

2022

Live radar dashboard reading a breathing rate of 33 and a heart rate of 61, with breathing and heart waveforms, chest displacement and a range profile

mm-Wave Radar-Based Vital Signs Monitoring and Arrhythmia Detection Using Machine Learning

Sensors (MDPI), 22(9):3106, 2022

2020

A crack in a railway track photographed on ballast, its extracted skeleton, and the skeleton overlaid back onto the crack in green

Structural Health Monitoring of Railway Tracks using IoT-based Multi-Robot System

Neural Computing and Applications, 33:5897–5915, 2020

2019

Grid of the Monument test image segmented by the WOA, ALO, FA, SSO and FASSO algorithms at four threshold levels

Antlion Optimization and Whale Optimization Algorithm for Multilevel Thresholding Segmentation

2019 Innovations in Power and Advanced Computing Technologies (i-PACT), IEEE

Projects

Talks & Demos

The ST Engineering team gathered around a humanoid robot at their exhibition booth, in front of a panel reading “From data to dexterity”

Robust Physical AI for UAVs and UGVs

2025

Four speakers in front of a screen announcing the IMDA Technical Sharing Session on Agentic AI, 30 September 2024

Agentic Apps: Lessons learnt from creating Agentic applications

2024

Visitors at the ST Engineering booth watching the AGIL AI Assistant demo on a large screen, under a sign reading “AI in Accelerating Sense-making”

AGIL AI Assistant: A multi-agent conversational chatbot

2024