Eswar Mahalingam · Robotics & Automation Intern · 10 simulation projects with from-scratch planning, control, SLAM, kinematics and computer-vision code; every number measured by this repo
Grid warehouse, differential-drive robot, ray-cast LiDAR, A* and Dijkstra planners, pure-pursuit, LiDAR-triggered re-planning, OpenCV item recognition, pick-and-place mission log.
| Picks / collisions | 6 / 6 · 0 |
| Driven vs planned | 147.2 / 150.3 m |
| Recogniser | 98 % |
| Tests | 17 / 17 |
Synthetic conveyor with 5 defect types, classical + ML defect detection, 4-DOF arm with analytic and DLS inverse kinematics, automated per-item inspection reports.
| Classical / SVM accuracy | 96.0 / 97.3 % |
| False alarms | 0 |
| Throughput | 21.5 items/min |
| Tests | 15 / 15 |
Smart-home device state machines, virtual robot, from-scratch MFCC + DTW keyword recogniser, intent/slot command parser with compound commands and error handling.
| Words @ 20 / 5 dB | 98.3 / 95.8 % |
| Spoken commands executed | 6 / 6 |
| Parser cases | 32 |
| Tests | 14 / 14 |
Nonlinear inverted-pendulum dynamics with motor model, IMU simulation, complementary vs Kalman fusion, cascaded PID, teleop commands, pole and gain-sweep analysis.
| Recovery from 10° | 0.65 s |
| Tilt RMSE (KF+enc) | 0.33° |
| Push 0.4 N·s → max tilt | 7.6° |
| Tests | 18 / 18 |
6-DOF quadrotor, WGS84→ENU GPS waypoints, geofence, return-to-home, rendered compound with moving intruders, MOG2 + HOG detection, tracking and alert engine.
| Waypoints reached | 12 / 12 |
| Intruders detected | 7 / 7 |
| Recall / precision | 97.7 / 99.5 % |
| Tests | 13 / 13 |
Field water-balance model, moisture/temperature sensor models, coverage rover, canopy image indices and leaf-spot detection, hysteresis irrigation with daily budget vs timer baseline.
| Water vs timer | −77 % |
| Stress-hours | 15 vs 266 |
| Health classifier | 96.0 % |
| Tests | 18 / 18 |
Skin-segmentation gesture recogniser (convexity defects, finger count, pointing), 6-DOF arm with Levenberg–Marquardt IK, One-Euro filtering and rate-limited joint trajectories.
| Recogniser (300 frames) | 95.6 % |
| IK converged | 200 / 200 |
| Jerk reduction | −97 % |
| Tests | 17 / 17 |
Rubble terrain, skid-steer with slip, 2-D LiDAR, log-odds SLAM with correlative scan matching + ICP, frontier exploration, colour + TDOA victim detection, mission map report.
| Victims found | 4 / 4 in 164 s |
| Coverage | 97.0 % |
| Pose RMSE SLAM / odom | 0.34 / 1.41 m |
| Tests | 10 / 10 |
3-level aisle graph, discrete-event simulator with peak arrivals, 4 slot-allocation strategies, bicycle-model manoeuvres, car-following and merge yielding, SAT collision monitor, ₹ dashboard.
| Collisions (16 runs) | 0 |
| Avg wait (tiered) | 53 s |
| Peak utilisation | 69 % |
| Tests | 12 / 12 |
Belt + camera renderer, colour/shape/Hu-moment classification with mm calibration, two SCARA arms with pick windows and EDF dispatch, throughput/OEE metrics and belt-speed sweep.
| Vision (600 items) | 99.7 % |
| Mis-sorts / missed (nominal) | 0 / 0 |
| OEE | 0.993 |
| Tests | 9 / 9 |