Large scale, diverse demonstration data for manipulation tasks remains a major challenge in learning-based robot policies. Existing in-the-wild data collection approaches often rely on vision-based pose estimation of hand-held grippers or gloves, which introduces an embodiment gap between the collection platform and the target robot. Teleoperation systems eliminate the embodiment gap, but are typically impractical to deploy outside the laboratory environment. We propose TRIP-Bag (Teleoperation, Recording, Intelligence in a Portable Bag), a portable, puppeteer-style teleoperation system fully contained within a commercial suitcase, as a practical solution for collecting high-fidelity manipulation data across varied settings. With a setup time of under five minutes and direct joint-to-joint teleoperation, TRIP-Bag enables rapid and reliable data collection in any environment. We validated TRIP-Bag's usability through experiments with non-expert users, showing that the system is intuitive and easy to operate. Furthermore, we confirmed the quality of the collected data by training benchmark manipulation policies, demonstrating its value as a practical resource for robot learning.
Figures & tables
Fig. 1 : The setup process of collecting data with our system. A few of the diverse locations we collected in.
In-the-Wild
Control
Action-Space
Direct Embodiment
Calibration
Operator Gripper
Full Proprioceptive
Interface
Representation
Mapping?
Free?
Feedback Mechanism
State Logging
UMI [ 6 ]
✓
Handheld
Task space
✗
✗
Passive trigger resistance
✗
ALOHA [ 1 ]
✗
Puppeteering
Joint space
✓
✓
None
✓
Gello [ 3 ]
✗
Puppeteering
Joint space
✓
✓
Passive spring mechanism
✓
Legato [ 11 ]
✓
Handheld
Task space
✗
✓
None
✗
DexCap [ 12 ]
✓
Handheld
Task space
✗
✗
Physical hand–object contact
✗
TABLE I : Comparison of Data Collection Devices
Fig. 2 : The breakdown of the hardware components and view of the cameras.
Submodule
Specification
Details
Suitcase
Weight
14.4 kg
Dimension
690 × 440 × 275 mm
Leader PC
Nvidia Jetson Orin Nano
Follower PC
Intel NUC10i3FNK
Wifi Router
Opal Wireless Travel Router
Power Supply
24 V / 15 A
TABLE II : System Specifications
Fig. 3 : Workspace of TRIP-Bag. All dimensions are in mm.
Fig. 4 : Overview of the software architecture, including device integration with the leader and follower PCs.
Fig. 5 : Range of deployment locations where the bag has been used for data collection and non-expert operation.
Fig. 6 : Step-by-Step Procedure for Target Tasks.
Fig. 7 : Set of interaction points from Task 2. Each colored cloud in the middle represents each set of the interaction points by categorization (For (a), based on location. For (b), based on users). Pictures shows the example interaction point from each set.
Fig. 8 : The number of successful episodes and average success time from the non-expert users.
Fig. 9 : Example Policy Rollouts for Task 1. Bottom row shows the predicted trajectory of end-effector from trained policy, calculated using forward kinematics models
Nowadays, robotic telemanipulation enables humans to control robots so as to execute tasks in a manner similar to how they perform them, and serves as an essential approach for collecting on-robot data for robot learning by human demonstrations. The quality of collected demonstrations directly impacts the performance of the learned policy. However, existing telemanipulation systems rely heavily on human operation, particularly for large-scale data collection. Collecting dozens or hundreds of demonstrations is a demanding process, as operators tend to fatigue quickly, especially during the execution of dexterous tasks that require sustained concentration. Recently, semi-autonomous robotic telemanipulation has emerged as a promising direction that reduces operator workload while preserving human-in-the-loop control for task execution. We therefore introduce a scalable, intuitive, semi-autonomous robotic telemanipulation system that leverages a mixed reality (MR) headset for on-robot data collection. The system autonomously executes the majority of the task trajectories and only requires the operator to make fine adjustments at key stages. We validate the effectiveness of our system by collecting data and training imitation learning policies on the execution of a high-precision, fine-grained bimanual peg-in-hole task. With lower data collection effort, our system obtains high-quality demonstrations that can be used to train fully autonomous task execution policies.
Shifei Duan, Zhe Wang, Francesco De Pace +1
New Dexterity research group, The University of Auckland, New Zealand · Competence Industry Manufacturing 4.0 (CIM4.0), Turin, Italy · National Technical University of Athens, Greece
Existing teleoperation systems are often tailored to specific robot hardware and task domains, limiting their scalability and adaptability. We present ModPack, a modular and extensible teleoperation system designed to support diverse robot embodiments and task requirements within a unified framework. At the core of ModPack is a self-contained wearable "backpack" that integrates onboard computation, power, communication, and data storage. Built on top of this shared interface, the system supports plug-and-play capability modules including joint-level teleoperation with haptic feedback, mobile manipulation, and active perception. Experiments across two distinct robot platforms and real-world mobile manipulation tasks demonstrate that ModPack provides a flexible and reusable framework for data collection and policy learning and collects higher quality data than a robot-free alternative. To support future research, we open-source the complete hardware design and software stack. Project website: https://modpack-robotics.github.io/
Behavior cloning for robot manipulation relies on expert demonstrations. However, for tasks that require dynamic stability, precise contact timing, or dexterous coordination, human operators may find it hard or even impossible to collect data. We study this infeasible-demonstration regime and propose GLIDE: Guardrails for Learning from Infeasible Demonstrations Efficiently, a framework that infers task-specific failure modes and converts them into executable guardrails for data collection and policy deployment. Given a task description and the conditioning teleoperation code, GLIDE writes guardrails that use system states to filter teleoperation and policy commands, constrain failure-prone actions, and iteratively improve from trajectory feedback. Across three tasks, GLIDE discovers emergent guardrails that go beyond domain-expert hardcoded ones, improving data collection over naive VR teleoperation and domain-expert hardcoded guardrails. After refinement, GLIDE raises data-collection success from 0-10 percent to 70-90 percent across the three tasks. During policy execution, mixed-data guarded policies reach 70 percent, 60 percent, and 60 percent success on Tomato plate transfer, Marker handover and stand, and Wine serving tasks. These results show that GLIDE can support policy learning when direct demonstrations are infeasible. Project website: http://guardrail-policy.github.io/