Harshvardhan
M.S. in Robotics | Seeking Fall 2026 PhD

Harshvardhan Chandirasekar

Researching the intersection of Human-Robot Interaction, Neural Rendering, and Mixed Reality to build intelligent systems that understand, collaborate, and learn from humans.

science View Research description Download CV
Currently at University of Minnesota Twin Cities
Harshvardhan Chandirasekar
person

Research Interests

Focusing on technologies that bridge humans and intelligent systems

blur_on

Neural Rendering

Interactive neural rendering with NeRF and Gaussian Splatting for AR/VR environments, enabling real-time 3D scene reconstruction with interactive capabilities.

  • Gaussian Splatting
  • NeRF
  • Real-time Rendering
view_in_ar

Spatial Computing

Human-computer interaction for spatial computing, building AR systems that enhance human capabilities in collaboration, training, and teleoperation.

  • Mixed Reality
  • Apple Vision Pro
  • Quest 3
smart_toy

Human-Robot Interaction

Human-robot interactions using mixed reality, enhancing teleoperation for everyday robot use through intuitive interfaces and natural interaction paradigms.

  • Teleoperation
  • LLM Integration
  • VLA Models

Experience

Research and industry experience

school
Current

Graduate Student Researcher

Zhu Tian Chen's Lab, University of Minnesota

Feb 2024 - Present

  • Built interactive Gaussian Splatting AR framework (90 FPS on Vision Pro)
  • Developed SpatialMiro - collaborative AR with LLM brainstorming
  • Trained uncertainty heads for VLA models (pi0)
  • Engineered low-latency GPU-CPU pipelines with FlatBuffers
memory

Research Engineer - Computer Vision

HiPeRT S.R.L, Modena, Italy

Sep 2022 - May 2023

  • Custom neural network on Hailo-8 & Jetson AGX (YOLOv7: 72.5% AP50, 60+ FPS)
  • Enhanced RTLS for automated forklifts with multi-device SLAM
  • WiFi-based warehouse localization (1.5m accuracy)
biotech

Research Intern (UG Thesis)

HiPeRT Lab, UNIMORE, Modena, Italy

May 2021 - Aug 2022

  • Real-time semantic SLAM - tkDNN + ORB-SLAM3 (24 FPS on AGX Xavier)
  • Reduced ATE 11.65%, APE 77.1%, ARE 74.2% on KITTI
  • Minimized semantic drift to 2 frames (vs baseline 10)

Projects

Selected research and engineering projects

music_note
JAX MediaPipe

Guitar Chord Recognition

Real-time chord recognition using hand pose estimation. Custom neural network in JAX with CNN and GNN implementations achieving 88% accuracy with ResNet.

code GitHub
sports_handball
Unity Quest 3

Learning Muscle Memory through AR

User study exploring accelerated muscle memory acquisition in AR environments. Custom physics system for controlled ball speed and gravity manipulation.

directions_car
RL CARLA

Emergency Response Automation

PPO-based reinforcement learning for autonomous vehicle navigation in CARLA with custom path planning for emergency rerouting.

robot_2
ROS LLM

Robot Navigation with LLMs

Integrated YOLOv7, whisper.cpp, and rosgpt to enable natural language robot manipulation commands with pose estimation on Baxter.

memory
CUDA TensorRT

tkDNN

Windows port and TensorRT 8+ compatibility. Custom CUDA kernels improving FP16 inference by 10% for YOLO networks. Support for transformer-based depth estimation.

code GitHub
security
AVR-BLE Swift

Home Security System

Real-time embedded security system with motion/door sensors on AVR-BLE. iOS app with secure element encryption and push notifications.

Get in Touch

Open to PhD opportunities and research collaborations

mail Email GitHub LinkedIn