Victor Gusev
Portrait of Victor Gusev

Specialized Robotics Engineer

Victor Gusev

2nd-year undergraduate at UCLA, on a three-year track. I build autonomous systems: reinforcement-learning planners, vision and LiDAR pipelines, and the embedded instrumentation underneath them.

Degree
B.S. Computer Science and Engineering
School
University of California, Los Angeles
Graduating
June 2028
GPA
3.64 / 4.00

Most recent

Software Engineering Intern @ Dimaag.ai

2026-062026-08

  • Architected a hybrid Reinforcement Learning and classical local planner for autonomous vehicles to improve accuracy and speed of travel along complex curved paths
  • Implemented PPO algorithm training pipelines, 2D LiDAR sensor raycasting, and domain randomization within NVIDIA Isaac Lab
  • Bridged the sim-to-real gap by validating policies across various simulators, implementing domain randomization to ensure seamless hardware deployment
  • Validated and deployed autonomous navigation policies on physical hardware, achieving reliable tracking at >12 mph with sub-decimeter trajectory accuracy
  • Integrated advanced LLM tooling into the engineering pipeline to automate architectural documentation, generate robust system test cases, and parse complex training logs

Previous Experience

Robotics Programming Lead · FIRST Robotics

2021-082025-05

  • Led software development for a 15-20 member robotics team, building semi-autonomous competition robots
  • Implemented PID control systems, vision tracking (Limelight), and LiDAR-based localization
  • Collaborated with mechanical and electrical leads to safely deploy and test software on the physical robot hardware
  • Engineered autonomous routines and vision pipelines that propelled the team to the FIRST World Championships, ranking in the top 5% of regional competitors

Outside the lab

Three things I do seriously, and what each one taught me that shows up in the work above.

Competitive paddling

Dragon Boat

500m erg
2:17
5k erg
18:50
Seat
Engine
  • Trains as an engine seat, the position responsible for sustained power output through the middle of the boat.
  • Working toward a sub-2:00 500m split by May 2027, with training programmed against stroke-rate targets for sprint, race, and head-race distances.

Structured training is a measurement problem before it is a fitness problem. Programming against stroke-rate bands and split targets is the same discipline as instrumenting a system: decide what to measure, hold the protocol steady, and let the numbers say whether the change worked.

Design for manufacture

Fabrication and CAD

Primary CAD
SolidWorks
Process
FDM
Shop
UCLA Makerspace
  • Designs rigid parametric parts in SolidWorks with explicit tolerancing, and prints them FDM at the UCLA Makerspace.
  • Reverse-engineers existing hardware with digital calipers to model mating parts against measured, rather than assumed, dimensions.

Designing a part that has to physically fit another part is unforgiving in a way software rarely is: a tolerance error is not a failing test, it is a printed object in the bin. It is the fastest feedback loop available for learning to specify something precisely before committing to it.

Formulas, by weight

Precision Baking

Measurement
By weight
Techniques
Sous-vide, fermentation
Built from it
Proof
  • Cooks and bakes to gram-accurate formulas rather than volume measures, treating a recipe as a parameterised process to be versioned and iterated.
  • Working through sous-vide, fermentation, and other technique-led methods where the controlled variable is time and temperature rather than ingredients.

Proof exists because of this. Keeping formulas by weight across dozens of iterations is a version-control problem wearing an apron, and the recipe PWA is the direct result of wanting real diffs between one bake and the next.

Projects

6 builds across software, robotics, and simulation.

softwarehardwarerobotics

Most recent build

Proof

Full-stack recipe PWA with AI-assisted import and real-time cross-device timers.