Shivansh Singh

A year building distributed autonomy software for a cargo aircraft in remote environments — part of North America's first peer-to-peer heterogeneous autonomous aircraft search-and-rescue mission (with the NRC). Worked across the flight stack: embedded drivers, real-time data systems, simulation, comms, and controls.

• Embedded C++17 drivers for safety-critical sensors and actuators — template metaprogramming for zero runtime overhead
• Real-time data path: zero-copy passing for deterministic sub-100 ms telemetry cycles
• Built the X-Plane 11 simulation bridge for hardware- and software-in-the-loop (HITL/SITL) testing of the full flight stack
• Integrated flight-software modules across state estimation and control into the autonomy pipeline
• Multi-threaded RF audio streaming over the comms link, with custom buffer management
• Defined test strategy and integration tests that surfaced integration issues on the ground, before flight

Built a pseudo-humanoid robot with computer vision, face recognition, and a ROS2 control stack — hardware and software from zero.

• 3D-printed chassis + custom PCBs wiring components to a Raspberry Pi 4
• ROS2 Humble on embedded Linux for modular robot control
• OpenCV computer vision in C++; PyTorch face-recognition models

Flight software for a hybrid VTOL UAV (multi-copter + fixed-wing). Two workstreams: a Modern C++ event-driven IPC framework for the autonomous flight nodes, and the PX4/Simulink flight-control stack.

• Architected a resilient event-driven IPC framework in Modern C++ — fault-tolerant crash resilience, asynchronous state synchronization across mission-critical cycles
• Defined the test strategy and integration test suite for flight-node protocols, reducing field-failure rates
• PX4 flight control for multi-copter/fixed-wing mode transitions; closed-loop attitude + rate controllers in Simulink; PID gain tuning from flight-log analysis

UofT's self-driving team. Led refactoring the deep-learning perception pipeline from Python into optimized C++ ROS2 nodes.

• Refactored latency-critical perception modules Python → C++ ROS2 nodes
• Profiled with Intel VTune + Valgrind; fixed cache misses and heap bottlenecks
• CUDA-accelerated inference on vehicle-grade GPUs; Dockerized CV pipelines on live vehicle hardware

2×+ throughput

~5× (memory-efficient ops) inference

~25% lower CPU at peak

Led UTRA's autonomous SUMO robot project — a combat robot that detects an opponent and drives it out of the ring while staying inside the edge itself.

• Managed the sub-team through design, build, and competition
• Coordinated the mechanical, electrical, and software workstreams — opponent/edge detection, motor control, and match strategy

Member of RSX's robotic-arm team, working on the inverse kinematics for a space-robotics manipulator.

• Worked on inverse-kinematics solving to position the arm's end-effector for manipulation tasks
• Mapped target poses to joint configurations for controlled, collision-aware arm motion

Worked on the test-track electronics for UofT's Hyperloop pod — sensing and control firmware on Arduino.

• Arduino firmware for the test-track sensing/control electronics
• Bring-up and testing of the electronics on the track hardware

UofT's student freelancing organization — built web applications for organization and client projects in React.

• React front-end development for an organization web platform