Skip to content
View Janga786's full-sized avatar

Block or report Janga786

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Janga786/README.md

Jangara Bliss

Computer engineering student at Fort Lewis College working on robot learning and embodied autonomy: benchmarks, simulation, controls, perception, and the embedded systems underneath them.

Portfolio: janga786.github.io · jangarabliss@gmail.com · LinkedIn

Research and flagship work

Project What it is
xembench Cross-embodiment, language-grounded manipulation benchmark in ManiSkill3: one policy interface, a Franka Panda arm and a Unitree G1 humanoid. 6,550 baseline episodes with Wilson CIs, a data-flywheel round, a documented negative result on precision manipulation, 198 tests
k1-vlm-navigation NaVILA vision-language model driving a velocity-tracking PPO locomotion policy on the Booster K1 humanoid in MuJoCo, with multi-step instructions, heading assist, and a real-robot deploy path (dry-run and live modes, watchdog, 58 unit tests)
k1_checkout_validator ROS 2 cart-seek stack for the real K1: YOLO-World detection, odom-frame tracking, and a gated, hard-clamped Booster SDK bridge with a lost-target stop
rebot-crack-vision Senior-design bridge crack-inspection arm (reBot B601-DM, 6+1 DoF): zero-shot crack segmentation, Jacobian/IK and minimum-jerk planning, and first powered hardware runs (Oct 2026): 7 motors over CAN at 200 Hz, gravity/friction feed-forward identified on the real arm, hold error ≤ 0.005 rad over 41 logged runs

Canonical NaVILA-on-K1 benchmark results (1,077 episodes, simulation) and the evaluation-reliability evidence behind the AIRC 2027 paper in preparation are in k1-navila-research; the write-up is on the portfolio.

Robotics, perception and controls

Project What it is
hexapod-cpg Custom 18-DoF hexapod from the NASA Colorado Robotics Challenge: Kuramoto-CPG locomotion, IMU heading-hold firmware, mechanical CAD, 34-test verification suite
kuka-kr6-kinematics From-scratch forward/inverse kinematics, geometric Jacobian, singularity diagnostics, and trajectory planning for the KUKA KR 6 R900, with a 39-test suite
lidar-pointcloud-motion-pipeline Open3D + NumPy point-cloud motion analysis: PCA orientation, frame-pair motion detection, ICP rotation-rate estimation
CV-YOLO-Inspection Synthetic-data inspection pipeline: Blender-rendered training data, YOLO11 detection, autoencoder anomaly detection
robot_vision Synthetic labeled training data from a single photogrammetry scan, using Blender domain randomization into YOLO training
Baxter-Sawyer_Work_Station_Windows Dockerized ROS 1 Indigo + ROS 2 Humble workstation with ros1_bridge for restoring and developing on Baxter/Sawyer

Embedded systems and hardware

Project What it is
basys3-fpga-portfolio Six Verilog systems on the Artix-7: PicoBlaze + OLED, XADC SPI bridge, PS/2-to-UART stack, soft-core CPU, FSMs
arduino-mega-microcontrollers Custom ATmega2560 PCB with production Gerbers, bare-metal 40 kHz ADC sampling, a hand-built IR link-layer protocol
cmos-vlsi-spice-portfolio Transistor-level CMOS design in 0.6 µm SCMOS: 8-bit ALU, R-2R DAC + flash ADC, transmission-gate MUX, full gate library

Software

Project What it is
cpp-algorithms-portfolio Six classical CS algorithms in self-contained C++17: union-find, MST, Huffman, optimal BST DP
basecamp Mobile-first training and habits PWA (React + TypeScript + Vite, swappable localStorage/Supabase backend, Google Health API sync)
pacman-pygame A complete Pac-Man clone in 443 lines of Pygame, with four target-chasing ghosts, power pellets, and a level state machine

Experience

  • Research Assistant, humanoid robotics (Fort Lewis College): integrated the Booster K1 into a vision-language navigation benchmark.
  • Senior design, bridge crack-inspection arm (Fort Lewis College, team of two): CAN motor bring-up, kinematics and trajectory planning, and controller tuning on the physical reBot B601-DM.
  • Team lead, NASA Colorado Robotics Challenge: led a four-person team that built and fielded an autonomous 18-DoF hexapod.
  • Research Assistant, AI and robotics (Fort Lewis College): restored legacy Sawyer and Baxter robots and built synthetic-data and inspection tooling.

Toolbox

Python · C/C++ · ROS 1/2 · PyTorch · MuJoCo · ManiSkill3 · Isaac Sim · OpenCV · Verilog · EAGLE/KiCad · Docker

Pinned Loading

  1. hexapod-cpg hexapod-cpg Public

    Custom 18-DoF hexapod robot — Kuramoto-CPG locomotion controller, real-time gait visualiser, IMU heading-hold autonomous firmware, full mechanical CAD, and a 34-test verification suite.

    Python 2

  2. k1-vlm-navigation k1-vlm-navigation Public

    NaVILA VLM drives a velocity-tracking PPO policy on the Booster K1 humanoid in MuJoCo: physical walking (not sliding) toward natural-language navigation goals, with multi-step instructions, heading…

    Python 2

  3. kuka-kr6-kinematics kuka-kr6-kinematics Public

    Complete kinematics suite for the KUKA KR 6 R900 sixx: from-scratch forward & inverse kinematics, geometric Jacobian, singularity diagnostics, trajectory planning, URDF generation, and a 39-test ve…

    Python

  4. xembench xembench Public

    X-Embodiment Language-Grounded Manipulation Benchmark: measuring language-conditioned policy transfer between a Panda arm and a Unitree G1 humanoid in ManiSkill3

    Python

  5. rebot-crack-vision rebot-crack-vision Public

    Bridge crack-inspection robot arm (reBot B601-DM): zero-shot crack segmentation, IK and trajectory planning, and CAN motor control with gravity/friction feed-forward identified on hardware

    Python

  6. k1-navila-research k1-navila-research Public

    NaVILA vision-language navigation on a Booster K1 humanoid: 1,077-episode VLN-CE-Isaac benchmark, RL locomotion policy, and an evaluation-reliability study (AIRC 2027, in prep)

    Python