Alex Shao

Researcher at Symbiokinetics

Building collaborative robotic platforms in healthcare at Symbiokinetics. Interested in Real-Time Control, Contact-Rich Manipulation, and Robot Learning. Pure Mathematics from Berkeley.

Portfolio of projects in embedded systems, GPU computing, and robotics RL.

Projects

• Robot Control-Library Architecture: End-to-end control architecture for the Franka FR3 and Emika Panda, including a 1 kHz PREEMPT_RT control loop, lock-free inter-process telemetry, a multi-machine offload layer, and a simulation backend. At Symbiokinetics.
• Generalized Compliant Control via Skill Discovery: Structured Unsupervised Skill Discovery (SUSD) combined with compliant control for auto-tuned, generalized motor behavior in simulation. At Symbiokinetics.
• High-Rate Custom Baremetal Firmware: Baremetal STM32 firmware for realtime data acquisition and streaming at high sample rates. At Symbiokinetics.
• Batched K-Means on GPU: GPU-accelerated batched K-Means clustering using CUDA C and Python for massively parallel computation. At Bongard AI.
• Simulating Backdrivability: Real-time rigid-body simulation for prototyping compliant control strategies. At Symbiokinetics.
• Pneumatic Soft Robot Manufacturing: Developed a repeatable heat-seal process for bonding LDPE pneumatic pouches to pressure valves. At Berkeley BMLab.
• Autonomous Sailboat Electronics: Designed and assembled electronics for a fleet of autonomous sailboats: IMU, magnetic encoders, wind vane, radio telemetry, and motor drivers. At Berkeley TAFLab.
• Handsi: Customizable desktop app mapping hand gestures to full OS actions: cursor, clicks, scrolling, volume, zoom, copy/paste, tab switching.
• TDS Publication: Batched K-Means. Towards Data Science publication with 10k+ reads.

Skills

Languages

• Python: Web dev, ML, and data since 2021. NumPy, PyTorch, MuJoCo, ROS2 bindings.
• C: Systems and embedded programming since 2023. ROS2, MuJoCo, Isaac Sim, POSIX, bare-metal.
• C++: Control systems and embedded since 2023. ROS2, MuJoCo, Isaac Sim, optimization.
• CUDA: GPU compute since 2023. Kernels, K-Means project, 1600x+ speedup.
• Rust: Systems programming since 2025. Handsi passion project.

Embedded and Systems

• STM32: Baremetal firmware. SPI, UART, I2C, SAI, DMA. LwIP, TCP/IP, mDNS. FreeRTOS.
• ROS2: Trajectory planning, joint control, sensor integration.
• EtherCAT: Real-time motor-drive control, deterministic fieldbus, and device state machines.
• Real-time Linux: PREEMPT_RT patched kernel, deterministic scheduling, sub-millisecond latency.
• Arduino: Autonomous sailboat electronics — IMU, encoders, telemetry, and propulsion/rudder/sail actuation.

Robotics and Simulation

• Isaac Sim and Isaac Lab: Robot modeling, domain randomization, sim-to-real transfer pipelines.
• MuJoCo: Physics simulation, contact-rich manipulation modeling, policy training and evaluation.
• libfranka (FR3 and Panda): 1 kHz torque and joint control of Franka FR3 and Emika Panda manipulators.
• MATLAB: Backdrivability and haptic interaction simulation, control prototyping and analysis.
• SolidWorks: Carbon-fiber end effectors, sensor mounts, and fixtures for Dexter HDI and Amber B1.

Machine Learning

• PyTorch: Tensors, autograd, custom modules, training loops, model deployment and inference.
• Reinforcement Learning: Policy gradient, off-policy, entropy-regularized methods.
• Imitation Learning: Behavioral cloning, residual policies, learning from demonstration.
• Skill Discovery: Unsupervised skill learning, density-based rewards, SAC (soft actor-critic).
• HPC: Batched K-Means on GPU, CUDA C and Python, 1600x+ speedup.