iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Yes—Harrison Kinsley (Sentdex) demonstrated a real robot picking up a toy from the floor and placing it in a bin without training a task-specific policy. The robot, Vulcan Robotics’ Sourccey (nicknamed “Chonk”), used camera images and a step-by-step prompt so a general-purpose multimodal model could plan the next action through the robot’s software development kit (SDK). The robot’s existing motion control, grippers and sensors carried out those decisions.
The result is best understood as a proof-of-concept for an image-guided planner, not evidence that an untrained language model has replaced robotic control. The fastest complete run reported was about six minutes (the demonstration video was sped up 10×), with roughly 90 model calls. No success rate, grasp-accuracy figure or repeated-trial benchmark was published.
What exactly did the robot do?
The demonstration used a staged sequence rather than one open-loop instruction. Kinsley prompted the model to inspect the scene, move the arm, verify what happened and continue only when the visual evidence supported the next step.
- Scout the workspace. Sourccey moved to a pose that gave its cameras a useful view of the floor.
- Search for the toy. The multimodal model interpreted camera images and identified the object to retrieve.
- Align an approach corridor. It positioned the arm so the toy sat in a workable part of the camera view and the gripper could approach without colliding with nearby geometry.
- Lower and grip. The arm descended and closed its gripper around the toy.
- Check the grasp. The model requested another observation to determine whether the toy was still in the gripper instead of assuming the action succeeded.
- Find the bin. It searched for the destination and selected a placement motion.
- Estimate distance with LiDAR. The robot used its range sensor to judge the bin’s position before depositing the toy.
- Release and finish. The gripper opened over the bin, completing the reported task.
This closed-loop pattern—act, observe, reassess and recover—is more significant than a single successful movement. It shows a model participating in a long-horizon physical task while the robot’s lower-level systems handled trajectory execution.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
Was an LLM really controlling the robot?
It was controlling the task at a high level, not directly generating motor currents. The model received visual observations and issued decisions through Sourccey’s SDK. Existing software translated those decisions into arm, wheel, gripper and sensor operations.
That division of labor matters. A language model can choose “move closer,” “lower the arm” or “verify the grasp,” but it does not automatically know the robot’s joint limits, collision envelope or camera-to-arm coordinate transform. Those capabilities came from the robot’s control and sensing stack, plus the prompts and movement guides Kinsley built.
Circuit Digest reported DeepSeek V4.1 Flash as the model supplying planning and control in the shown workflow. Kinsley also said Z.ai GLM-5.3-Flash could perform the task. The report does not establish that either model is universally capable of this behavior, nor that the models were fine-tuned on toy-picking demonstrations.
How did the multimodal model know where to move the gripper?
It combined visual interpretation with a calibrated, incremental procedure. Rather than asking for a complete six-axis trajectory from one image, the prompt presented a small decision at each stage and then supplied a new observation after movement.
Image-guided alignment
The model looked for the toy in the camera frame and guided the arm toward an approach corridor. This reduces the problem from “solve the entire room” to “put the target in a position where the next motion is safe and useful.”
Verification after contact
After closing the gripper, the workflow explicitly checked whether the object remained held. That check is essential because a visually plausible closing motion can still miss, slip or catch only part of an object.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
Range sensing at the destination
LiDAR supplied a distance estimate for the bin. Camera images provide appearance and bearing, but a range measurement can help distinguish a nearby opening from a similarly sized object farther away. Sourccey’s LiDAR is a 360-degree 2D unit, so it contributes a horizontal scan rather than a full three-dimensional model.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →These steps do not mean the model inferred perfect metric geometry from pixels. They show a planner using available observations and robot primitives to make progress while repeatedly checking the result.
What robot did Sentdex use?
The platform was Vulcan Robotics’ Sourccey, which Kinsley calls “Chonk.” Vulcan describes it as an open-source developer robot for real-world data acquisition and physical AI. Its published 2026 specifications list the following hardware:
| Component | Published specification | Why it matters in the demo |
|---|---|---|
| Mobility | Four mecanum wheels | Allows the base to translate and rotate in a constrained workspace while positioning the arms. |
| Manipulation | Two arms, each with five degrees of freedom and a gripper | Provides the joints and end effectors used to reach and hold the toy. |
| Onboard compute | Raspberry Pi 5 | Runs the robot-side software and SDK integration; the model’s inference role is described separately from low-level actuation. |
| Vision | U20CAM-720p cameras; five cameras are discussed in the calibration report | Supplies views for scouting, alignment and verification. |
| Range sensing | 360-degree 2D LiDAR, up to 12 m on white objects, configurable 5–13 Hz scan rate | Provides distance information used when locating the bin. |
| Battery | 12 V, 10 Ah LiFePO4 pack | Published power specification for the mobile platform. |
| Size and mass | 1,030 mm tall; 15.88 kg | Sets the physical scale and handling requirements for a floor-level task. |
The manufacturer says the robot is serviceable: servos are individually addressable and replaceable, and public STL files support replacement parts. That openness is useful for experiments because a damaged actuator or printed bracket need not make the entire platform disposable.
Why was calibration the hardest part?
Calibration was the main engineering bottleneck because the robot’s sensors do not share one fixed viewpoint. Circuit Digest describes five cameras and a 2D LiDAR mounted at different angles. Cameras attached to the grippers move as the arm joints move, changing the relationship between pixels and the robot’s base every time the arm changes pose.
Moving camera geometry
A pixel location that corresponded to the toy from one arm pose will not map to the same physical ray after the camera moves. The system therefore needs a per-camera movement guide or equivalent kinematic model so image observations remain meaningful as joints change.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
Coordinate-frame errors
Small errors in camera extrinsics, joint zero positions or gripper offsets can produce a large miss at floor level. A planner may correctly identify the toy while still sending the gripper beside it if the coordinate transform is wrong.
Simulation and iteration
Kinsley built a per-camera movement guide and used simulation while iterating on GLM-5.3-Flash. He still reported calibration issues that required fixing. That experience illustrates a general rule: multimodal reasoning cannot compensate for an uncharacterized sensor frame.
How long did the task take, and how many model calls were used?
The fastest complete run reported by Circuit Digest in 2026 took about six minutes. The video was sped up 10×, so playback time is not a direct measure of physical latency. The workflow made roughly 90 model calls per task.
Recommended Free Tools
Those numbers describe one reported demonstration, not an average. No published data gives the number of attempts, the distribution of completion times, the cost per call or how often the robot recovered from a failed grasp. A high call count also means network or inference delays could matter in a less controlled environment.
Is this a trained robot policy or an LLM calling the SDK?
For the toy task, it was an LLM-driven planner calling the SDK, not a documented task-specific vision-language-action policy trained on a toy-picking dataset. The robot’s existing motion and sensing routines executed the selected actions, while the model chose what to do next from observations and prompt instructions.
| Dimension | Prompt-driven multimodal planner in this demo | Task-trained vision-language-action policy |
|---|---|---|
| Training data | No task-specific training was reported; behavior came from prompting and robot-side software. | Usually requires demonstrations, labeled trajectories or reinforcement-learning data for the target tasks. |
| Control authority | High-level SDK decisions; established controllers execute motion. | May output lower-level actions or trajectories, depending on the policy design. |
| Recovery behavior | Prompt explicitly included visual checks and another decision after contact. | Depends on whether recovery examples and state estimation were included during training. |
| Latency and calls | About 90 model calls were reported for one task; timing varies with inference and communication. | Can run at a fixed control rate once deployed, but training and deployment compute requirements vary. |
| Calibration | Still requires accurate camera, LiDAR and joint geometry. | Also requires calibration; a trained policy does not remove coordinate-frame errors. |
| Reproducibility | Prompt, SDK version, model availability and sensor setup all affect results. | Weights, data, hardware and control stack must be held constant to reproduce results. |
| Evidence in this report | One demonstrated sequence; no success-rate or accuracy benchmark. | Would normally be evaluated over repeated trials and standardized metrics. |
The comparison is not a claim that one approach is always better. Prompting can make a new task accessible without collecting a specialized dataset, while a trained policy may offer more predictable timing once its operating envelope is established.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
What does “open source” mean for Sourccey?
Vulcan presents Sourccey as a platform for learning robotics and physical AI, with support for LeRobot-compatible teleoperation, recording, training and rollout workflows. The official build guide covers the base, chassis, wiring, Raspberry Pi setup, arm construction, sensor mounting, actuator and potentiometer setup, calibration and safety checks.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minutePublic hardware files include CAD for the body, arms, dome, linear-actuator parts, accessories, teleoperator and URDF assemblies. Open files can make repair and experimentation easier, but they do not eliminate the need for mechanical assembly, electrical troubleshooting, calibration and safe test procedures.
What would you need to reproduce the experiment?
A faithful reproduction requires more than a multimodal API key. At minimum, you would need:
- A Sourccey platform with its arms, grippers, cameras, LiDAR and Raspberry Pi 5 compute.
- The robot SDK and a software path for sending model decisions to safe, bounded motion primitives.
- Accurate camera, joint and LiDAR calibration, including transforms for cameras that move with the arms.
- A prompt or state machine that separates scouting, approach, grasp verification, bin localization and release.
- A controlled test area with a known toy and bin, clear of people and loose hazards.
- Logging for every image, model response, command, sensor reading and recovery decision.
There is no published cost, official kit price or verified success rate for this particular setup in the reported material. Component prices and model-inference charges can change, so a bill of materials would not reliably predict the cost of duplicating Kinsley’s configuration.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What are the main failure modes?
The demonstration exposes several ways a system like this can fail. Treating them as ordinary software bugs would be unsafe because an incorrect command moves physical hardware.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Mislocalized toy: the model selects a visual distractor or estimates the toy’s center incorrectly. Use additional views and require confidence or a human-confirmed target during early tests.
- Transform drift: the camera-to-arm relationship is wrong after a joint moves. Recalibrate joint zero positions and validate the transform with a known calibration target.
- Missed grasp: the gripper closes beside or over only part of the object. Keep the post-grasp image check and command a retreat-and-retry branch rather than proceeding to the bin.
- Occluded bin: the destination is hidden or confused with another opening. Change the scouting pose and use LiDAR consistency checks before release.
- Unsafe command: a generated action exceeds a joint, speed or workspace limit. Put hard limits in the SDK layer so the model cannot bypass them.
- Communication or inference delay: a stale image leads to a decision about a scene that has changed. Timestamp observations and reject commands based on outdated frames.
Human supervision remains prudent while calibrating or changing prompts. A stop control, bounded velocities and a test object that cannot injure someone are practical safeguards.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
What does this demonstration prove—and what does it not?
It demonstrates that a general-purpose multimodal model can participate in a multi-step physical task on a real, sensor-equipped robot without a reported toy-specific training phase. The sequence included visual search, arm positioning, grasp verification, destination finding and LiDAR-assisted placement.
It does not establish a production-ready autonomous home robot. The available report supplies no success rate, grasp accuracy, repeated-trial reliability, end-to-end latency benchmark or cost analysis. The six-minute run and approximately 90 calls should therefore be read as measurements from a reported run, not guarantees for other toys, rooms, models or software versions.
The broader lesson is architectural: a capable multimodal model can serve as a flexible task planner when paired with a well-instrumented robot, but reliable physical action still depends on calibration, constrained controllers and explicit recovery checks.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchFrequently Asked Questions
Can the same prompt make Sourccey handle any household object?
Not on the evidence available. The reported run involved one toy and a bin in a controlled setup; different shapes, materials, lighting and clutter would change both perception and grasping difficulty.
Does the robot need an internet connection for this approach?
The report identifies cloud-style multimodal model calls but does not specify the deployment location or network architecture. In practice, any remote inference path would add dependence on connectivity and variable response time, while a local model would require sufficient compute and compatible software.
Why use both cameras and LiDAR instead of cameras alone?
Cameras provide color and appearance cues for recognizing the toy and bin. The 2D LiDAR contributes a direct horizontal distance measurement, which can help with placement geometry even though it does not produce a full 3D reconstruction.
Can I evaluate reliability from the six-minute video?
No. A video shows that at least one run completed, but it does not reveal failed attempts, the total trial count or the probability of success. Those require a repeated-trial protocol with predefined success criteria.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
