Book a call ↗THE TECH STORYTELLER / BUILD IN PUBLIC
All 15 published episodes, with YouTube links and individual field notes. Plus the sensor story across episodes and Heron’s design films.
FOLLOW THE EXPERIMENT
ONEBOX / DESIGN + REAL SENSOR TESTS
Design revisions, chip choices and actual prototype output: watch the bring-up, calibration and bent-wall test.
Explore the Onebox build →
DESIGN EXPERIMENT / 3 VERTICAL FILMS
Heron’s body studies, robot-arm neck, tactile hand and the gap between a render and a walking robot.
Explore Heron & watch the films →Editorial episode covers include AI-assisted illustrations. Episode 5 uses an original reconstruction still.
Publication check: 22 October 2026. EP01–EP15 are public. New experiments in production are added after release.
Watch on YouTube ↗EPISODE 15 · PUBLISHED
The iToF sensor reports a straight wall bowed 12–14 cm. Longer exposures reduce noise, receiver bugs are patched and multiple causes investigated, yet frequencies disagree by 22 cm. A one-wall fitted correction is local only; cyclic error and spill remain hypotheses, while TMF8829 dToF is considered for a wrist sensor.
2026-10-21
Read the episode field note →
Watch on YouTube ↗EPISODE 14 · PUBLISHED
An under-$100 RGB + ToF + IMU concept is evaluated with a custom simulator, then an epc660-class engineering sample is brought up beside a Pi camera. A cable causes two hours of brown-outs. Calibration improves 12.7 to 4.56 pixels but misses the <4 px target; a 30-second map motivates a shared clock, IMU and factory calibration.
2026-10-20
Read the episode field note →
Watch on YouTube ↗EPISODE 13 · PUBLISHED
A 52 cm Roadster-inspired CO2-thruster concept moves from CAD to simulation. Ideal hover gives way to gas-flow limits, valve restrictions, cooling and real component mass. Seven-second hover is an ideal simulation result, with no physical gas test performed.
2026-10-16
Read the episode field note →
Watch on YouTube ↗EPISODE 12 · PUBLISHED
A Pi Zero 2 W takes control of a toy robot over Bluetooth. Twenty-one SDK tests pass without movement; captured app traffic leads to packet timing fixes, a restart and roughly four seconds of neutral initialization. The physical sequence works but mismatched wheels make forward spin; autonomy remains future work.
2026-10-15
Read the episode field note →
Watch on YouTube ↗EPISODE 11 · PUBLISHED
Gemini gets cameras, LiDAR and code-writing access to a simulated ECO65 basketball setup. After fixing the desk, physics and gripper, the final five-throw test scores four baskets, while Gemini mislabels one made shot. This is a small fixed simulation test.
2026-10-14
Read the episode field note →
Watch on YouTube ↗EPISODE 10 · PUBLISHED
The live OS0 is brought back online and localized in the earlier room map. Walking through the iPhone view drives three calibration iterations; a convincing wall fit initially hides a wrong camera pose. Actual body point clusters become a rough avatar, with estimated intrinsics and unseen anatomy.
2026-10-13
Read the episode field note →
Watch on YouTube ↗EPISODE 09 · PUBLISHED
The office twin receives simulated Go2 and G1 robots. Calibration, movement, scene reconstruction, lighting and grasping failures precede a humanoid fall when the arm is forced overhead. All robot results are simulation.
2026-10-12
Read the episode field note →
Watch on YouTube ↗EPISODE 08 · PUBLISHED
A real Ouster room scan and 13-second phone video become a simulated physical-AI dojo with splats, labelled objects, colliders and movable parts. Alignment and robot training infrastructure remain experimental; physical parameters are estimates.
2026-10-11
Read the episode field note →
Watch on YouTube ↗EPISODE 07 · PUBLISHED
GPT-6 co-engineers a low-cost quadruped carrying two SO-101 arms. Across 38 recorded simulation runs, 19 fail; slipping blocks, vision, connection loss and tipping shape the design. A deterministic controller executes the demonstrated run.
2026-10-10
Read the episode field note →
Watch on YouTube ↗EPISODE 06 · PUBLISHED
The planned RealMan ECO65 setup is explored using manufacturer CAD and robot description: cameras, LiDAR, interfaces and a target-pose control loop. The hardware has not yet arrived; all arm sequences are simulation.
2026-10-09
Read the episode field note →
Watch on YouTube ↗EPISODE 05 · PUBLISHED
A 48.6-second phone video produces a room reconstruction without LiDAR input: 187/195 selected images align, 51,475 sparse points and 400,000 Gaussians. Appearance works, while metric scale and unseen surfaces remain unverified.
2026-10-08
Read the episode field note →
Watch on YouTube ↗EPISODE 04 · PUBLISHED
Live Ouster data reaches an old Samsung S10 through a Mac bridge. Pose jumps, IMU compatibility and a disappearing Ethernet adapter expose the gap between displaying scans and reliable mapping; phone localization is still future work.
2026-10-07
Read the episode field note →
Watch on YouTube ↗EPISODE 03 · PUBLISHED
A Raspberry Pi camera and Ouster OS0 produce a coloured point-cloud experiment. Checkerboard and manual matching improve selected static boundaries from about 13.5 to 4.2 pixels, but the person ghost remains unresolved.
2026-10-06
Read the episode field note →
Watch on YouTube ↗EPISODE 02 · PUBLISHED
An old Samsung S10 maps the office and gets a simulated robot body. Route replay and virtual exploration work; an attempted autonomous pickup remains unfinished.
2026-10-05
Read the episode field note →
Watch on YouTube ↗EPISODE 01 · PUBLISHED
Mechanical engineering meets general-purpose AI controlling robot arms: a commentary on a source dual-arm pick-and-place demonstration, introducing the question of what actually works.
2026-10-01
Read the episode field note →HERON / 9:16 DESIGN FILMS
The current launch, hardware and reinforcement-learning explainers. Blender renders and MuJoCo simulation, with my synthetic voice.
9:16 · 2:06 · DESIGN / SIMULATION
The latest launch cut: the arm-neck, beak hand and three proposed use cases. Includes two labelled shots from the first robot version.
Open video ↗9:16 · 1:59 · DESIGN / SIMULATION
The motor layout, tactile hand, sensor head, electronics and exploded assembly. Contact grasps are simulated.
Open video ↗9:16 · 1:46 · DESIGN / SIMULATION
An RL explainer covering tasks, reward, domain randomisation and sim-to-real. Walking and falls are scripted illustrations.
Open video ↗Current walking is scripted; contact grasps are simulated. These are design films, not footage of a completed physical robot. Read the design story and limitations →
SHOW THE WORK / WATCH THE MOMENT
Short excerpts from the published episodes. Each keeps the result in its original scope.
▶ The robot-dog side quest
▶ The Physical AI dojo
▶ The LiDAR / camera experiment
▶ An old phone, a virtual robot
▶ Why physical AI caught my attentionThe origin Short includes commentary, third-party excerpts and AI-assisted presentation. Episode excerpts below use the original published edit.
WHAT THE FORMAT LOOKS LIKE
Independent editorial examples. These projects were not paid for by, sponsored by or affiliated with the featured manufacturers. They illustrate the format of a possible collaboration, not an existing client relationship.
Ouster OS0 · calibration, mapping & sensor comparisons
Read the case study →RealMan ECO65 · digital twin, basketball & Heron
Read the case study →Unitree Go2 / G1 · support, grasping & balance in simulation
Read the case study →Phone video + LiDAR · from capture to robot tasks
Read the case study →
Design revisions · chip choices · real bring-up recordings
Explore the build →LET’S BUILD A USEFUL STORY
Bring a product, a workflow or an engineering question. We agree the experiment, deliverables, publication and usage before the work starts.
Discuss a collaboration↗Pricing on request · lukas@alphavant.com