Advanced Robotics and AI intern at Siemens. MS Robotics at Northeastern University ('27), BE in Computer Science from Osmania University. Before grad school I spent five years on Team Robocon MJCET, working up from junior programmer to leading a 20+ member team to its first national competition in five years.
I work on physical AI and the motion stack underneath it — planning, reactive replanning, legged locomotion, and the unglamorous bringup that makes any of it run on real hardware. Lately I've been pointing learned policies at deformable objects and watching them fail in instructive ways.
| Now | Advanced Robotics and AI intern at Siemens, Berkeley |
| Building | Sim-to-real transfer of bimanual garment policies onto a UR12e |
| Next | Inverse RL on quadruped locomotion, aimed at a Unitree Go2 |
| Always | Planners written from scratch, then measured rather than assumed |
| System | What it does | Numbers |
|---|---|---|
|
Bimanual garment manipulation
|
Trained and evaluated ACT, Diffusion Policy and smolVLA on multi-garment folding in a bimanual setup, alongside work with π0, π0.5 and OpenPI across the VLA stack. Sim-to-real transfer onto a UR12e is in progress. |
60.4% success |
|
Reactive replanning on a UR12e
|
UR12e + Robotiq Hand-E. A depth pipeline with workspace filtering, robot self-masking and colour-based body exclusion isolates live obstacles; the arm cancels cleanly and re-plans from its current state mid-motion. |
100% detection |
|
drive it · race the controllers
|
A modular planner/controller architecture built from scratch — no Nav2 BT server. A*, Theta*, SMAC and RRT behind one global interface; DWA, Pure Pursuit, Stanley, TEB and MPPI behind another. |
99–192× faster |
|
Robot foundation model
|
A flow-matching VLA, a latent world model and an RL-tuned reasoner on one backbone, with an orchestration layer above that never touches the weights. It trains, and it has still never completed the task. The one real result is the competence signal. |
0 of 16 completed |
|
|
A quadruped with locomotion modelled after the MIT Cheetah. Gait stability and autonomous navigation validated with SLAM in Gazebo; now collecting demonstrations for inverse RL. |
12 axes |
|
|
DWA written from scratch rather than pulling in |
6 stars |
Not a screenshot. A GitHub Action seeds a fresh map every night, fetches
astar_planner.pyandtheta_star_planner.pyfrom the nav repo, runs both searches, and renders the expansion order into the plate above. The node counts and millisecond figures are whatever that run produced — the generator is here. Both planners get the same map and the same seed, so the difference in how much of it they had to look at is the difference between the two algorithms.Want to drive one yourself? These repos run in your browser — search a map you draw with any of five planners, chase a cursor with the DWA controller while you throw obstacles at it, race all five local controllers over one plan, or reach into the UR12e's cell and block the arm until it cancels the motion and replans around you.
Where these numbers come from
Counted from the git history of every repo on this account. Commits are filtered to my own
authorship; the language mix excludes two repos built on forked upstreams (K.A.L.B,
lehome-challenge) so vendored code doesn't skew the split.
| Language | Share | Bytes |
|---|---|---|
| Python | 30.7% | 594 KiB |
| CMake | 18.1% | 351 KiB |
| YAML | 15.0% | 290 KiB |
| XML / URDF / XACRO | 14.8% | 286 KiB |
| C++ | 8.5% | 164 KiB |
| Other (C, Shell, Arduino, HTML) | 12.9% | 249 KiB |
| Total | 100% | 1.89 MiB |
| Commit cadence | |
|---|---|
| Commits authored | 240 |
| Window | Jul 2021 → Jul 2026 (61 months) |
| Active months | 20 |
| Median, active month | 8 |
| Peak month | 49 (Apr 2026) |
2026 ──● Siemens · Advanced Robotics and AI Intern Jun 2026 –
│ Berkeley, California. On-site.
│
● Northeastern University · Graduate Lab Assistant Apr – Jun 2026
│ ROS 2 bag collection from a Ford Mustang Mach-E autonomous
│ vehicle platform, with visual SLAM run over the recorded drives
│ for trajectory analysis. UR12e URDF with Robotiq integration and
│ the MoveIt 2 stack for faculty research, plus an end-effector
│ camera mount now used across several setups.
│
2025 ──● MS Robotics, Electrical & Computer Engineering 2025 – 2027
│ Northeastern University
│
2020 ──● BE Computer Science & Engineering 2020 – 2024
│ Osmania University, Hyderabad
│
● Team Robocon MJCET Dec 2020 – Jul 2025
│ Junior Programmer → Programmer → Robotics Control Engineer
│ → Team Lead/Captain → Team Mentor
│
│ ABU Robocon 2021 · pitch-pot holonomic chassis, pneumatic
│ arrow handling, spring-based shooter
│ ABU Robocon 2022 · lagori omni shooter + swerve, CV for
│ incoming balls, ML predicting where the stack would fall
│ ABU Robocon 2023 · Angkor Wat omni + mecanum, ring and pole
│ detection, collection and shooting
│ ABU Robocon 2024 · Harvest Day the team's first national
│ appearance in over five years. Two holonomic drives, ball
│ shooters, pneumatic seedling collection, silo placement. ROS
│ waypoint navigation, colour-based ball detection, and an ML
│ policy that picked silo placements against the opponent.
│ ABU Robocon 2025 · robot basketball 2v2, as mentor — ROS guidance
│ and play strategy
│
2022 ──● Consciente Technologies · Robotics Engineer Intern Aug – Sep 2022
│ DH parameters and MoveIt configuration for a custom 5-DOF
● manipulator; coordinate-based autonomous navigation on hardware.
Send a goal pose
building robots that may work · if it planned on the first try, the obstacle was imaginary