Work Experience – Hridaanshu (Anshu) Gusain

Overview

Built a lab-ready 4‑DOF robotic arm and mobile base that merge embedded control, perception, and simulation for rapid mechatronics R&D. The system serves as a proving ground for motion‑planning ideas before hardware deployment. It is in development for deployment at the Molecule Maker Lab to autonomously navigate around the laboratory floor, interacting with scientists and test vials.

Concept

Use a compact manipulator paired with an NVIDIA Orin and depth camera to prototype autonomy quickly. A shared codebase in ROS2 lets the same behaviors run in Gazebo and on the real robot.

Design Process

• Modeled kinematics/URDF and stood up a full simulation loop in ROS2 + Gazebo for CI‑like testing.
• Implemented PID/state‑machine control on STM32/ESP32 with watchdogs and safety interlocks.
• Integrated depth sensing for mapping and state estimation; validated SLAM + tag localization workflows.
• Closed the sim‑to‑real gap with log‑driven tuning and bench HALT/soak tests on actuators and gripper.

Specifications

Results

• ~40% faster iteration by validating trajectories in simulation first.
• Improved localization accuracy with fused vision + IMU; reduced drift on long runs.
• Bench fault rate reduced via watchdogs and safety state machine.

Overview

Around 50% of bees are affected by varroa mites, a parasitic mite that can cause significant damage to bee colonies. These infections are costly and can lead to colony collapse. MiteOut builds low‑power, field‑ready data collection and monitoring tools for precision commercial beekeepers. I co‑founded the effort and lead the embedded hardware/firmware stack.

Concept

Deploy ESP32-based nodes with calibrated sensors and reliable RF to monitor hive health continuously. Focus on robust power, OTA serviceability, and modular boards for quick pilot spins. Reliability and longevity is of the utmost importance, becuase these decices have to perform year-round in the open environment.

Design Process

• Designed 4‑layer mixed‑signal PCBs (KiCad/Altium) with protected power stages and low‑noise analog front‑ends.
• Wrote FreeRTOS firmware in C++ for task scheduling, sensor pipelines, and OTA diagnostics.
• Ran environmental soak tests and field telemetry to guide board iterations and firmware fixes.
• Set up Python tooling for calibration and automated log analysis.

Specifications

Results

• Increased field uptime by ~50% through RTOS scheduling and power management.
• Cut validation time by ~30% with automated Python/C++ calibration suite.
• Secured $8K in pre‑seed funding and deployed pilot units to multiple apiaries.
• Currently deployed in commercial and hobbyist beehives across Upstate New York, collecting long‑term field data.

Overview

A 6‑DOF Robotic Arm for Advanced 3D Printing. The NovoPrint project (ASME • Dr. Clemon’s group) aims to redefine additive manufacturing by printing on non‑planar surfaces and complex geometries. The system integrates mechanics, electronics, and software control into a cohesive robotic platform based on the open-source platform ARCTOS.

Concept

Adapt a robust open‑source arm and tailor it for precision additive manufacturing. By mounting a printing extruder as the end‑effector and leveraging the dexterity of a 6‑axis robot, we unlock toolpaths impossible on traditional 3‑axis gantries.

Design Process

• Led cross‑functional work across mechanical design, electronics, and motion control; coordinated funding and milestones.
• Fabricated and assembled extensive custom parts and fixtures; integrated CAN‑synchronized stepper subsystems.
• Implemented ROS motion control, precision calibration, and a serial command interface for bring‑up.
• Validated non‑planar printing with feedback; iterated via rapid prototyping and bench testing.

Specifications

Results

• Successful non‑planar toolpaths and synchronized multi‑axis prints with reduced drift.
• Awarded UIUC EOH Distinguished Technology (2025).
• Reduced print time by ~20% through multi‑extruder synchronization.

Overview

Currently, about 50% of cancer biopsies are misdiagnosed due to the difficulty of identifying cancerous cells. Current technologies implement computer vision techniques to identify cancerous morphologies. However, they are often inaccureate with around <60% accuracy. I was the lead researcher (under the mentorship of Dr. Mestha) studying deep learning pipelines with hyperspectral nuclei segmentation techniques to improve accuracy to around 85%.

Concept

Apply Hyperspectral imaging techniques using the UNET model architecture to high‑dimensional biomedical imagery; tighten the loop between dataset prep, training, and validation. Improve accuracy, and validate on different image stacks.

Design Process

• Built TensorFlow training stack; engineered preprocessing with NumPy/pandas.
• Constructed datasets and threshold‑tuning scripts; tracked experiments for comparability.
• Collaborated on writing and figures as IEEE first author.

Specifications

Results

• Accuracy improved from ~50% → ~85%; presented internationally.
• IEEE Systems Council Honorable Mention (2024).

Overview

Lead Programmer for FRC 2791, delivering competition‑ready control, vision, and autonomous behaviors.

Concept

Engineer a robust software stack in Java/C++ for swerve drive, vision‑assisted scoring, and reliable autonomous routines.

Design Process

• Implemented motion profiling and real‑time PID/PIDF tuning.
• Integrated fiducials/AprilTags for pose estimation and autonomous alignment.
• Led a 25+ member software team with reviews, training, and task delegation.

Specifications

Results

• Improved autonomous scoring reliability; qualified for 2024 Worlds (Hudson Valley #3 seed).