Agilex COBOT Magic (COBOT Magic)

Agilex COBOT Magic (COBOT Magic)

COBOT Magic is positioned as a research-oriented, ready-to-integrate platform for universities, laboratories, and industrial R&D teams that need a repeatable hardware stack for experiments in mobile manipulation, imitation learning, and robotic telepresence. The system’s feature set emphasizes practical data workflows (collection, replay, visualization, training, inference) and integration with a ROS-based software ecosystem.In typical deployments, COBOT Magic is used to record demonstrations of real-world tasks such as grasping, pick-and-place, and bimanual manipulation then replay, visualize, and train models on the collected data for autonomous or semi-autonomous operation.

Design and Features

Mobile base architecture

COBOT Magic is built on a differential-drive mobile base associated with AgileX’s TRACER chassis family. Product materials describe a two-wheel differential drive configuration with a payload class intended to carry manipulation hardware and compute on top of a compact platform footprint. 

Multi-arm manipulation layout

A defining feature of COBOT Magic is its multi-arm configuration. AgileX describes the hardware as being based on four robotic arms. 
This layout aligns with common Mobile ALOHA-style setups, where “leader” (teaching) arms can be used for demonstration while “follower” arms execute motions on the robot, supporting bimanual and whole-body teleoperation data collection.

Vision and sensing

COBOT Magic is typically configured with a top-mounted camera plus wrist cameras, supporting egocentric (hand-centric) and scene-level views for manipulation learning. AgileX specifies two wrist cameras and one top camera.
A referenced specification sheet lists an Orbbec depth camera with an example depth range of 0.3–3 m, including RGB and depth streaming modes intended for perception and dataset capture. 

Software focus: data collection and training workflows

AgileX positions COBOT Magic as an integrated environment for (1) setting up the Mobile ALOHA software stack, (2) collecting demonstrations, and (3) training models across simulated and real-world environments. 
Public community posts about COBOT Magic also emphasize an end-to-end pipeline covering data collection, replay/visualization, training, and inference.

Technology and Specifications

Mobility and power

AgileX lists COBOT Magic with a headline payload of 100 kg, runtime around 4 hours, and top speed around 1.6 m/s. 
A published spec sheet also describes a compact chassis footprint and a charging time around 4 hours for the base. 

Manipulator characteristics

A reference specification sheet describes a 6-DOF arm class with a rated payload of about 1.5 kg (with higher peak load figures in short bursts), reflecting the platform’s emphasis on manipulation research rather than heavy industrial payload handling. 
Separately, distributor materials commonly describe 6-DOF arms and research-grade repeatability targets for teleoperation and learning experiments. 

Compute and robotics stack

A published spec sheet references an NVIDIA Jetson Orin Nano (8GB) developer kit for data acquisition in at least one configuration. 
For software, the same document lists Ubuntu 18.04 LTS and ROS Melodic in its example OS stack, along with common robotics tools such as Gazebo/MoveIt/RViz-style supervisory visualization and simulation workflows. 

Applications and Use Cases

Academic and industrial research in mobile manipulation

COBOT Magic is frequently presented as a practical platform for mobile manipulation research, where navigation and manipulation must be solved together (e.g., moving through cluttered indoor spaces and then interacting with objects). AgileX and exhibitors describe tasks ranging from simple grasping to more complex sequences such as pouring and organizing objects. 

Teleoperation and demonstration learning (imitation learning)

A primary workflow is teleoperated task execution to produce datasets: operators demonstrate tasks, sensors capture multi-view video and robot state, and that data becomes training input for learned policies. This is especially relevant for projects aiming to scale real-world skill learning without hand-coding behaviors. 

Simulation-to-real experimentation

AgileX highlights the ability to set up training across simulation and real environments, supporting experiments where models are prototyped in simulation and validated on the physical robot. 

Pilot deployments and applied R&D

Beyond academia, such platforms can be used for early-stage trials in warehouse labs, corporate innovation centers, and robotics integrator testbeds—particularly where the goal is evaluating feasibility of learned manipulation for repetitive indoor tasks rather than immediate production automation. 

Advantages / Benefits

COBOT Magic is commonly selected for four practical reasons:

End-to-end learning workflow orientation. It is marketed not only as hardware, but as a system designed around demonstration capture, replay, training, and inference loops. 

Mobile + bimanual manipulation in one platform. The combination of an indoor mobile base with multi-arm manipulation enables research on tasks that require both repositioning and dexterous interaction (a core challenge in embodied intelligence). 

Multi-view perception for manipulation. Wrist cameras plus a top camera support data capture that is well-aligned with modern learning-based perception pipelines.

Open-source/ROS ecosystem alignment. Published materials and resources emphasize a ROS-based architecture and supporting packages/resources, which can reduce integration time for robotics teams already working in ROS.

FAQ Section

What is AgileX COBOT Magic?

AgileX COBOT Magic is an open-source mobile manipulation robot platform that combines a differential-drive base, multi-arm manipulation hardware, and cameras to support teleoperation, dataset collection, and embodied AI research. 

How does COBOT Magic work?

COBOT Magic is commonly used by teleoperating the robot to perform tasks while recording sensor streams and robot state. Those demonstrations can be replayed and used to train models for learned task execution, using workflows aligned with Mobile ALOHA-style data collection and training pipelines. 

Why is COBOT Magic important?

It provides a standardized way to study mobile manipulation—a key problem in robotics—by unifying navigation, bimanual manipulation, and multi-view perception in a system designed for data-driven learning and experimentation. 

What are the benefits of COBOT Magic for robotics teams?

Key benefits include a learning-oriented workflow (data capture → replay → training → inference), a mobile base paired with manipulation hardware for end-to-end experiments, and ROS-aligned resources that can speed up integration for research and R&D teams. 

Summary

AgileX COBOT Magic is a research-focused mobile manipulation platform built to support teleoperation-driven data collection and embodied AI training. By combining a TRACER-class differential mobile base, multi-arm manipulation hardware, and multi-view camera sensing within a ROS-aligned ecosystem, it targets the practical needs of labs and R&D teams working on bimanual mobile manipulation, imitation learning, and simulation-to-real experimentation.