A Backpack That Controls Robots
Most teleoperation systems are built for one robot, one task, one lab. Joshua Citron and his team at the University of Washington are trying to change that with ModPack, a modular teleoperation interface designed to work across different robot embodiments and task requirements within a single, wearable framework.
The system centers on a self-contained backpack unit that houses onboard computation, power, communication, and data storage. From this shared backbone, users can attach capability modules as needed: joint-level teleoperation with haptic feedback, mobile manipulation, or active perception. The idea is to turn teleoperation from a custom integration project into a plug-and-play experience.
Hardware Agnosticism by Design
The researchers tested ModPack across two distinct robot platforms and a range of real-world mobile manipulation tasks. The results, detailed in their arXiv paper, show that the framework can collect training data and support policy learning without requiring platform-specific re-engineering for each new robot. This is significant because data collection remains one of the biggest bottlenecks in scaling robot learning. If a single operator interface can work across arms, mobile bases, and end-effectors, the cost of generating diverse demonstration datasets drops sharply.
The backpack form factor is deliberate. By putting the compute and power on the operator rather than tethering to a nearby workstation, ModPack enables movement through real environments. That mobility matters for tasks that require the operator to navigate a space while controlling a robot, such as fetching objects from different rooms or manipulating items on a mobile base.
Open Sourcing the Stack
To accelerate adoption, the team has open-sourced both the complete hardware design and the software stack. This follows a growing trend in robotics research, where reproducibility and community extension are prioritized over one-off demonstrations. For labs that lack the resources to build custom teleoperation rigs from scratch, ModPack offers a starting point that can be adapted rather than rebuilt.
The broader context here is the push toward generalist robot policies. Models like RT-2 and OpenVLA have shown that large, diverse datasets can produce surprisingly capable robot controllers. But those datasets require human demonstrations, and human demonstrations require interfaces. ModPack positions itself as infrastructure for that pipeline: a standardized, extensible way to put humans in the loop so that robots can learn from them at scale.
From Lab to Field
The team's experiments focused on bimanual mobile manipulation, one of the harder teleoperation scenarios because it requires coordinating two arms while potentially moving a base. ModPack's modular approach means a researcher can start with a single-arm setup, add a second arm later, and eventually integrate mobility, all using the same core backpack and software architecture. That incremental expandability is rare in a field where systems are often rebuilt from the ground up for each new degree of freedom.
For the robotics community, the release of both hardware schematics and software represents a practical contribution that could lower the barrier to entry for teleoperation research and data collection. Whether it becomes a standard platform will depend on how well the community adopts and extends it, but the design philosophy, prioritize modularity and openness over optimization for a single platform, aligns with where the field is heading.