Dobot X-trainer

The DOBOT X-Trainer is an advanced AI data collection and robot training system, specifically developed for the requirements of modern AI research, robotics education, and industrial development projects. At its core are two Dobot Nova 2 cobot arms, combined on a compact platform with a global camera, master-slave controller, and powerful software. This creates a system that bridges real-world motion data with AI-based modeling, enabling complex tasks to be demonstrated by humans and subsequently executed autonomously by the robot.

With its diverse functions for data collection and model training, the DOBOT X-Trainer allows for the rapid acquisition of new skills and the seamless execution of complex instructions. Typical scenarios range from dishwashing or laundry folding to stocking shelves, as well as assembly and material handling in industrial environments. Thanks to its high precision and robust mechanics, the system is suitable for both laboratory setups and practical production and service testing.

Dual Nova 2 Cobot Arms with High Precision

The system is based on two Dobot Nova 2 cobots. These six-axis collaborative robot arms each offer a 2 kg payload, a reach of 625 mm, and a repeatability of ±0.05 mm. In the X-Trainer, they are combined to create a large, overlapping workspace: in dual-arm mode, the working area can be extended up to 1200 mm, allowing both fine-motor and large-scale tasks to be covered reliably.

The industrial precision of the Nova 2 arms ensures that data and motion trajectories captured in educational or research environments can later be transferred directly to real-world automation applications. Processes such as pick-and-place, screwing, sorting, or laboratory handling can be trained in the protected environment of a teaching or research lab and subsequently transferred to production robot cells without having to re-develop the underlying trajectories or policies.

Rapid Data Collection and Training through Imitation Learning

A key feature of the DOBOT X-Trainer is the efficient combination of teleoperation and imitation learning. The Nova 2 cobot arms are controlled directly via the master hand controllers. Movements, gripping processes, and interactions with the environment are continuously recorded and converted into datasets for subsequent training. Routine scenarios such as washing dishes, folding textiles, or sorting objects can thus be fully learned in about two hours—reducing training time by up to 70% compared to classic programming approaches.

The 25 Hz high-performance motion interface ensures that the robot follows hand movements with almost no delay. This not only makes teleoperation more comfortable, but also increases the quality of the recorded data: smooth transitions, natural accelerations, and finely dosed forces are captured cleanly and made available to the imitation learning models. This makes the X-Trainer excellent for research projects involving imitation learning, offline RL, or the training of policies for complex manipulation tasks.

Safe Human-Robot Collaboration and Ergonomic Operation

Since the DOBOT X-Trainer is primarily used in laboratories, training rooms, and development environments, safety plays a central role. The underlying Nova 2 cobots feature 5-level collision detection and ISO 15066 certification, ensuring that forces are limited in the event of a collision and that cooperation with humans can take place safely. Additionally, the workspace is padded with flexible, shock-absorbing EVA foam to minimize damage to grippers, tools, or workpieces in case of accidental drops.

Special safety logic engages in the event of malfunctions or error states: in the event of a power failure or alarm, the master and slave arms are automatically desynchronized, and detection of the master hand being dropped prevents uncontrolled movements. If necessary, the system resynchronizes the master and slave at a safe, low speed to avoid collisions. Together with the ergonomic controllers, which weigh less than 400 g, the result is a system suitable for both long teleoperation sessions and classroom situations with varying users.

Areas of Application in Education, Research, and Industry

The DOBOT X-Trainer is designed for a wide range of applications. In educational institutions such as schools, universities, and vocational colleges, it enables practical courses on topics such as robotics programming, AI-based automation, teleoperation, and data processing. Students learn to set up real-world scenarios, perform demonstrations manually, collect data, and derive models from it, which then independently control robot tasks.

In research institutes and AI labs, the X-Trainer serves as an Embodied AI platform for AGI scene simulation and experimental AI algorithms. Through the precise control of the dual arms, the large workspace, and the open API, a wide variety of tasks can be represented: from manipulating household objects and experimental manufacturing tasks to complex multi-step scenarios where the robot must react to disturbances and adapt its strategies.

The X-Trainer also demonstrates its strengths in industry-related development environments. Companies can test new automation steps as prototypes without immediately building complete robot cells. Typical examples include learning handling and assembly steps that are later transferred to stationary cobots, or generating large datasets for AI-assisted quality inspection, grasp planning, or trajectory optimization. Thanks to the identical Nova 2 kinematics, the transition from the lab platform to production applications is comparatively simple.

Open Interfaces and Complete SDK

Included with the DOBOT X-Trainer is a complete SDK that maps the entire pipeline from data collection and model training to autonomous inference. Developers have access to joint angles, forces, RGB and depth data from the cameras, and control commands for grippers and tools. This data can be integrated into proprietary AI frameworks or exported in formats directly suitable for domain-specific neural networks (e.g., ALOHA-like architectures).

This makes the X-Trainer not only a teaching and demonstration system, but a serious tool for data-driven robotics development. Whether for imitation learning, reinforcement learning, policy optimization, or multimodal perception models, the platform provides high-quality raw data and a stable hardware base upon which various research directions can be built.