GATE MAP

The languages and physics every later phase stands on: C++, ROS 2, rigid-body kinematics, circuits, control refresh.

• C++ to intermediate level (classes, RAII, templates, CMake)
• ROS 2 basics: nodes, topics, services, custom messages, launch files
• Modern Robotics ch. 2–6: configuration space → forward/inverse kinematics
• Electronics bench basics: read schematics, breadboard, multimeter, ESP32 blinky → motor PWM
• Control refresh: transfer functions, PID, root locus, Bode (Åström & Murray)
• BOSSROS 2 package that drives a simulated 2-link arm with your own FK/IK

Both robots alive in simulation + real firmware on real silicon. New PC lands mid-phase.

• MuJoCo humanoid: load a humanoid model, PD joint control, standing balance controller
• PX4 SITL beyond drone_inspector: offboard mode, custom flight modes, mission scripting
• State estimation: Kalman filter → EKF, implement attitude estimator from raw IMU data
• Embedded: STM32 bare-metal (GPIO, timers, PWM, I2C/SPI drivers) — FastBit course
• First PCB: design + order a simple sensor breakout in KiCad
• BOSShumanoid stands under pushes in MuJoCo + drone flies scripted survey in SITL

Train the robots. Deep learning properly, then RL, then a walking policy. Drone parts ordering begins (₹50k budget unlocks).

• PyTorch fluency: training loops, CNNs, deployment (ONNX) — Bourke course + Karpathy Zero-to-Hero
• RL fundamentals: Sutton & Barto pt. 1, implement DQN + PPO from scratch on Gym tasks
• Humanoid locomotion: PPO walking policy in MuJoCo (CPU-trainable; new PC accelerates it)
• Drone perception: vision-guided precision landing on a marker in SITL
• Order drone hardware: frame, Pixhawk 6C, motors/ESCs, Pi 5 companion, camera (~₹40k)
• BOSShumanoid walking video + drone vision-landing in sim, both on the build log

The first physical robot: assemble, wire, tune, then fly an autonomous vision mission outdoors (DGCA rules: sub-2kg, registered, open category).

• Assemble + wire the quad: soldering, power distribution, ESC calibration
• Bench bring-up: sensor calibration, PID tuning, first manual hover
• Companion computer: Pi 5 + camera running your perception stack, MAVLink offboard
• Autonomous mission: takeoff → waypoint survey → vision-guided precision landing, no pilot input
• BOSSfull flight demo video + build writeup: your CAD, your wiring, your code, your trained model

Close the loop on the second robot: transfer the walking policy from MuJoCo to a real mini-biped (budget round 2).

• Choose + acquire platform: mini-biped / small humanoid (serial-bus servos, ~13–17 DOF)
• System ID: match sim model to real hardware (masses, friction, servo response)
• Domain randomization retrain: policy robust to sim-real gap
• Deploy: policy running onboard, robot walks untethered
• BOSSwalking robot video + full 15-month retrospective post