A new laboratory often starts with a shopping list. A better starting point is an experiment: what must the robot sense, control and repeat? This shortlist covers ten platforms already represented in the directory. The sequence groups experimental needs; it is not a benchmark ranking, a price comparison or a claim that one platform is best for every lab.
Define the experiment before the hardware
For manipulation, specify the objects, contact conditions, workspace and control interface. For navigation, write down the floor surface, available positioning signals and sensor mounting needs. For whole-body learning, define the support equipment, reset procedure and supervision needed. A visually impressive demonstration is not a substitute for access to the signals your experiment requires.
Choose the smallest complete experimental system
An arm is only part of a manipulation setup. Include a gripper, cameras, fixtures, workstation, calibration tools and a safe operating area. A mobile base may need sensors, batteries and time synchronization. Ask whether the quoted configuration includes the development interface, example code, support and replacement parts; a consumer configuration may not expose the same interfaces as a research version.
Test reproducibility and recovery
Ask for a sample log and run a small repeatable task before committing to a platform. Record observations, actions, timestamps, software versions and calibration. Measure how long a student takes to recover from a failed grasp, lost localization or protective stop. A robot that produces usable data reliably can be more valuable than one with a more ambitious demonstration.
A practical first-month acceptance plan
Week one: assemble the system and reproduce an official example. Week two: log observations and commands for a simple task. Week three: repeat with different objects or start positions. Week four: have a second researcher reproduce the setup from written instructions. These are proposed acceptance steps, not measured outcomes for the candidates below.
Resource shortlist
Ordered by workflow, not by measured performance. These are editorial starting points. Follow each directory entry to its official documentation and confirm the exact configuration before purchasing.
Franka Research 3 →
A starting point for contact-rich manipulation and force-sensitive robot-learning experiments.
Check before committing: Required control modes, real-time workstation setup, gripper and current development-interface terms.
Kinova Gen3 →
A candidate for arm-based manipulation research where integration and accessible control interfaces matter.
Check before committing: Exact arm variant, software interface, gripper compatibility and support for your chosen middleware.
UFACTORY xArm →
A candidate for tabletop manipulation and development with an arm-and-tool setup.
Check before committing: Payload at your tool offset, SDK behavior, calibration and the supplied gripper configuration.
Interbotix Arms →
A route into tabletop manipulation and teaching with the Trossen development ecosystem.
Check before committing: Specific arm model and lifecycle: Trossen reports legacy hardware end-of-life. Confirm current availability, replacement parts and maintained software before procurement.
TurtleBot3 →
A compact platform for indoor navigation, mapping and ROS-oriented teaching experiments.
Check before committing: Chosen model, computer and sensor package, ROS distribution and battery support.
Clearpath Jackal →
A ground-platform candidate for autonomy research beyond a small indoor tabletop setup.
Check before committing: Terrain assumptions, sensor integration, payload mounting and the quoted development package.
Clearpath Husky →
A candidate for ground robotics with larger sensors and field-oriented integration needs.
Check before committing: Actual terrain, transport requirements, weather constraints and power budget.
Hello Robot Stretch →
A mobile-manipulation candidate for studying reaching and interaction in human environments.
Check before committing: Workspace, object requirements, software stack and whether the experiment needs additional sensing.
ALOHA →
A bimanual research platform for demonstration collection and imitation-learning experiments.
Check before committing: The original project is a research reference. Trossen reports its legacy ALOHA hardware line is end-of-life; distinguish a reproduced research build from current Trossen AI products and confirm parts/support.
Unitree G1 →
A whole-body research candidate when the study specifically requires a humanoid form.
Check before committing: Research/developer configuration, accessible interfaces, fall management and supervised operating procedures.
Turn the shortlist into a decision
Write down the task, operating conditions, success criteria and total ownership costs. Ask shortlisted suppliers to demonstrate your representative task, including failure recovery. Use the same acceptance test for each candidate.
Explore the full robotics directory →