// user.profile —— online
Shiyao Ni
M.Sc. Mechanical Engineering @ McGill & Mila · advised by
Prof. Audrey Sedal.
Currently Embedded AI Intern at Aerial Technologies. I build
AI for hardware — both physics-accelerated design
of soft systems and embedded intelligence running on the edge.
About
I’m a graduate researcher at McGill University and Mila — Quebec AI Institute, working with Prof. Audrey Sedal on embodied intelligence — specifically RL for robotic grasping and manipulation across soft and rigid grippers, and acoustic tactile sensors that give soft hands a sense of touch. I care about agents whose bodies are part of how they think: morphology and control, sensing and actuation, all co-designed. My undergraduate research at McGill led to a publication in Autonomous Robots on sim-to-real transfer of co-optimized soft crawlers — a project that pulled me deep into computational design and convinced me the trial-and-error loop of physical design can be replaced by something much smarter.
Before grad school, I worked on physics-accelerated design as an ML consultant at Mila — coupling diffusion models and LLM-guided search with deterministic physics simulators to replace blind RL search over soft-robot morphologies (50% faster convergence than RL baselines). Earlier I shipped closed-loop control of agricultural robots at TopxGun’s CTO Office, and co-optimization of crawling soft robots with domain randomization as an undergraduate researcher. Right now I’m an Embedded AI Intern at Aerial Technologies, putting AI back on the hardware: on-device classifiers (fire alarm, fall detection) on ESP32-S3 with real-time piezo / PVDF signal conditioning.
Selected work
RL for grasping & tactile sensing
M.Sc. thesis: simulation-trained RL policies deployed via ROS on open-source arms; acoustic tactile sensors integrated onto soft grippers for grasp-state sensing.
Physics-accelerated design for soft robots
Diffusion models and LLM-guided search coupled with deterministic physics simulators — replacing blind RL morphology search with physics-informed automation. 50% faster convergence than RL baselines.
Design work & side projects
Mechanical design, embedded prototypes, and personal builds across hardware and software.
What I'm thinking about
RL and generative models for materials and design spaces — making the trial-and-error cycle of complex design faster and more principled.
Publications
Sim-to-real transfer of co-optimized soft robot crawlers (Autonomous Robots, 2023) — and more in prep.
Latest resume
Full resume in PDF — research, projects, internships, coursework, and the toolbox.