Robotics Researcher
I build Vision-Language-Action models for robot navigation and autonomous driving.
Robotics Engineering, University of Michigan · Ann Arbor, MI
My work brings classical state estimation together with deep learning — from SLAM and sensor fusion to language-conditioned policies — to make autonomy that holds up in the real world.
I am a Robotics Engineering student at the University of Michigan, focused on perception, navigation, and learning for autonomous systems. I am currently researching navigation models at the Scalable Spatial Intelligence Lab.
My interests lie in path planning, deep learning, SLAM, computer vision, and probabilistic representations of the world. I am drawn to problems where robots need to understand complex, uncertain environments and build structured internal models that support reliable decision-making — how machines represent space, reason under uncertainty, and connect perception to action in a principled way.
I am especially interested in approaches that bridge geometric methods with learning-based systems, combining classical state estimation with deep neural networks to create scalable, robust autonomy stacks where mapping, localization, and visual understanding are core intelligence layers rather than isolated components.
Long term, I want to contribute to autonomous systems that perceive richly, learn efficiently, and operate reliably in the real world.
Language-conditioned policies that perceive, reason, and act — for navigation and autonomous driving.
Nift · DreamerVLA · CARLA SFF
Learning navigation models and policies that generalize to unseen, uncertain environments.
Scalable Spatial Intelligence Lab · ObjectNav BC
Probabilistic localization and mapping — invariant EKF, particle filters, and multi-sensor fusion.
Full-Stack AV (IEKF) · MBot SLAM
Sampling-based and search planning under real-time and uncertainty constraints.
FRoG Lab · Ocean Frontier (NBVP)
B.E. in Robotics Engineering — GPA 3.96 / 4.0
Honors & Awards: Ranked 3rd, World Robotics Olympiad · 4× medalist, International Youth Robotics Contest · 3× AMC qualifier · 2× AIME qualifier
Selected coursework: Control of Robotic Systems (ROB 415), Planning & ROS Integration (ROB 320), Localization & Mapping (ROB 330), Data Structures & Algorithms (EECS 281), SLAM, Robot Optimization, Linear Algebra, Robot Differential Equations
Autonomous Vehicles — Vision-Language-Action Models
Undergraduate Researcher — Navigation Models · University of Michigan
Undergraduate Researcher — Perception & Planning · University of Michigan
Undergraduate Researcher · University of Michigan
Embedded Software Engineer · Ann Arbor, MI
Instructional Aide · University of Michigan
Instructional Aide · University of Michigan
Instructional Aide · University of Michigan
Selected work across embodied AI, autonomous driving, SLAM, and multi-robot systems — from research pipelines to hardware that shipped.
ResNet18 + GRU · PIRLNav · Habitat
Trained a ResNet18 + GRU navigation policy from 76K+ PIRLNav human VR demonstrations across 80 HM3D scenes. Implemented truncated BPTT, 3.28× inflection weighting, and a custom dataset loader to bypass Habitat's strict episode schema — an end-to-end embodied-AI pipeline from raw demonstrations to a deployable navigation policy.
Exploration & Planning
Autonomous three-phase ROV mission: BFS sweep of unmapped voxels, RRT-biased centroid approach, and full NBVP survey. Detects wrecks via vertical column clustering; fuses stereo depth and sonar into a real-time OctoMap.
AV Safety
A Dockerized CARLA / ROS 2 closed-loop evaluation stack for the Alpamayo VLA. Designed 10+ long-tail scenarios and refactored the safety layer into a formal Safety Force Field (Nister et al., 2019), benchmarked with unit tests.
Perception
Used event-camera microsecond temporal resolution and high dynamic range to detect low-observable satellite streaks against star-field clutter, with asynchronous spike-pattern processing for space domain awareness.
SLAM
Full SLAM on the MBot platform: laser-based localization, occupancy-grid mapping, and autonomous exploration using frontier detection and path planning.
Localization
An autonomous-vehicle localization pipeline using IEKF sensor fusion (IMU, wheel odometry, GNSS, stereo/mono VO, LiDAR), with map-based corrections for drift-bounded pose tracking. Read the full write-up.
Learning & Navigation
A vision-language-action (VLA) model for robot navigation and control, exploring how language-conditioned policies can drive embodied agents toward goals. Code on GitHub.
Planning & Mapping
Generalized planning and mapping for microrobot systems, developed in the Synergic Adaptive Machines Lab to coordinate motion and reconstruct the environment across many small agents.
Robotics Systems
A C++ HTTP + serial bridge and browser UI controlling a 9-motor array with real-time feedback and preset motion patterns.
Dynamics
MATLAB dynamic simulation of a humanoid leg, modeling joint torques, ground contact, and human-like gait.
Embedded
An accelerometer-based robotic limb with precise real-time finger control and motion sensing.
Sensing
Black-ice detection robot using sensor fusion; alerts pedestrians with LEDs and an audio module.
Robotics
Solar-powered robot that waters plants on highway medians using a robotic arm.
Energy
A dual energy-harvesting system drawing from pressure and heat to power EV batteries autonomously.
Automation
An agricultural robot that ploughs, sows, and waters, accelerating indoor crop growth with artificial light.
Developer Tools
A Visual Studio Code extension that streamlines ROS 2 development workflows inside the editor.
Manipulation
A robotic arm that can see, learn, and act to manipulate a variety of objects.
Swarm Robotics
A multi-robot warehouse logistics system using swarm robotics for coordinated task allocation.
// the stack I work in
ashrith@umich:~$ cat languages.txt
C++ · Python · Julia · MATLAB
ashrith@umich:~$ cat robotics.txt
ROS 2 · RViz · Gazebo / Isaac Sim · tf2 · rosbag · STM32 · Arduino · I²C · sensor fusion
ashrith@umich:~$ cat control_perception.txt
LQR · PID · particle & Bayes filters · Kalman filtering · SLAM · A* / D* · RRT · occupancy mapping
ashrith@umich:~$ cat math_design.txt
linear algebra · ODEs/PDEs · optimization · least-squares · Jacobians · SolidWorks · kinematics
ashrith@umich:~$ cat dev_tools.txt
Linux · Git · CMake · gdb · profiling · reproducible builds
ashrith@umich:~$ ros2 topic list
/cmd_vel · /scan · /tf · /odom · /recruiter/callback (1 subscriber, finally)
ashrith@umich:~$
Developed a piezoelectric energy-harvesting system converting automobile pressure into electricity to power electric vehicles.
1 Gold, 2 Silver, 1 Bronze
2× USAMO qualifier, 3× AIME qualifier
Qualified for the USAMO (top 300 of ~300,000 globally) and earned AIME distinction across three years.
I am always glad to connect with fellow researchers, collaborators, and curious minds. Reach out and let's build something together.