Can phones and tablets connect without a local router?

Yes. The camera automatically broadcasts a Wi-Fi hotspot (reCamera-Arm-Teleop, IP 192.168.16.1). Connecting to this network allows opening the web cockpit on port 8765 immediately.

Which robotic arm models are verified?

The hardware is verified on the reBot B601-DM arm with 7 Damiao dual-mode motors. Drivers for serial bus servos (SO-ARM101) exist in code but require specific hardware verification.

Smart Robot Arm with Vision Brain

Give the robot arm a vision brain — no industrial PC needed, drive it from your phone, or let AI agents take over.

A single reCamera serves as both eyes and brain for the reBot arm. Mount it directly on the arm to drive 7 joint motors over CAN with no external industrial PC. Drive it immediately from a phone browser, or let AI agents take full closed-loop control over native tools.

SenseCraftIntermediate30minVision & Video AIAI Agent ReadySource Code

Reusable Site Templates

An all-in-one embodied manipulation platform eliminating bulky on-site industrial PCs.

Router-Free Wireless Teleoperation — Connect mobile phones directly to the camera AP hotspot to jog joints or adjust Cartesian poses from the browser.
Embodied AI Teaching & Trajectory Collection — Support gravity compensation drag-teaching, recording 50 Hz synchronized joint episodes into Parquet files.
Host-Assisted Agent Manipulation — Edge TPU detects targets and ArUco 6D poses; external AI agents control closed-loop actions via rebot-companion MCP tools (pick-and-place verified in simulation).

How it works

A standard reCamera (SG2002 RISC-V SoC) paired with the B4 CAN base board acts as the arm controller and wrist vision sensor. It runs armd for 50 Hz joint control and web cockpit, and rebot-vision for H.264 streaming and TPU YOLO / CPU ArUco inference. External AI agents interact via 14 native MCP tools.

Zero-Setup AP Hotspot — Broadcasts reCamera-Arm-Teleop Wi-Fi hotspot (192.168.16.1) for direct mobile phone cockpit access without external routers.
Native MCP Protocol — Exposes 14 tools (get_state, snapshot, detect, move_to, gripper, grasp, etc.) to Claude Desktop and Claude Code.
Demonstration Recording — Records synchronized joint telemetry and commands at 50 Hz directly into Parquet files with gravity compensation dragging.

reCamera (SG2002) + B4 CAN Board — Embedded RISC-V controller running armd. Static binary is 4.1 MB, RAM footprint 3.7 MB, single-core CPU usage 7.0 % at 50 Hz.
On-Demand TPU YOLO — Runs YOLOv11 target detection in 47 ms per frame on hardware TPU.
CPU ArUco 3D Pose — Detects 4x4_50 ArUco markers and estimates 6D pose in real time on CPU.
Interactive Eye-in-Hand Calibration — In-browser guided calibration using a single printed A4 marker; exports transformation matrix JSON for companion picking.

reBot B601-DM Arm — 6 joints and 1 force-controlled parallel gripper, 7 Damiao servo motors driven over 1 Mbit/s SocketCAN (can0).
50 Hz Closed-Loop Motion — Real-time position, velocity, and torque control with forward and inverse kinematics.
Safety Lease Watchdog — Client commands must hold active leases. Automatic motion halt and safe retreat if network drops.

How well it works

Measured on reCamera (SG2002) with B4 CAN base board + reBot B601-DM (Damiao motors):

Metric	Measured Value	Hardware / Role
`armd` daemon binary size	4.1 MB	Static binary on RISC-V Linux
`armd` runtime memory	3.7 MB	RSS RAM usage
Closed-loop CPU usage (7 motors @ 50 Hz)	7.0 %	Single SG2002 CPU core
Edge TPU YOLOv11 inference	47 ms / frame	On-chip TPU (34 % chip load at 5 FPS)
CPU ArUco marker pose estimation	37–57 ms	CPU single thread (640×360)
Demonstration capture sampling	50 Hz	1509 frames recorded in 30 seconds

What you get out of it

Interfaces exposed by reCamera (:8765, :8001, :8090):

Interface	Protocol	Purpose
`http://<camera-ip>:8765`	HTTP / HTML	Multi-touch web cockpit (joysticks, Cartesian jog, calibration)
`ws://<camera-ip>:8765/api/v1/ws`	WebSocket	Bidirectional arm control channel (telemetry, commands, lease)
`ws://<camera-ip>:8001/`	WebSocket / H.264	Real-time video stream from wrist camera (`rebot-vision`)
`http://<camera-ip>:8090`	HTTP / REST	On-demand vision pipeline (YOLO detections & ArUco 3D poses)
`rebot-mcp`	MCP (stdio)	14 tools exposed to Claude Code and Claude Desktop

Porting It to Your Own System

Modular components of the reCamera + reBot architecture:

Layer	Modular Interface	Porting / Modification
Vision Pipeline	`rebot-vision`	Swap YOLO weights in `/userdata/rebot-teleop/models` or replace detector
Actuator Bus	`armd` SocketCAN	Adapt URDF kinematics and motor CAN protocol profiles
Agent Integration	`rebot-mcp`	Add domain-specific prompt tools or integrate into custom agent loops

How to deploy

Check the setup and its product list, then follow the deployment guide.

Hardware Sizing & Deployment Setup

Select site conditions to generate recommended bundles and guides, or let an AI assistant estimate hardware and deploy for you.

Self-service Consult & Deploy

Deployment Options

reCamera + reBot B601-DM

Install the rebot-teleop app on the camera and the AI agent server (MCP) on your computer, then check teleop and run hand-eye calibration in the cockpit.

3 steps · ~30min

Deployment Guide →

AI Camera

recamera_2002

Robotic Arm

rebot_arm_b601

FAQ

No. The reCamera (SG2002 RISC-V SoC) paired with the B4 CAN base board runs the full control daemon and drives the 7 Damiao motors directly over SocketCAN.

Didn't find your answer?

We Are Glad to Be Your Hardware Partner !