What Mobile Manipulation Unlocks
Fixed-base manipulation robots are limited to tasks within their workspace — a volume of 1–3 cubic meters centered at a fixed point. The vast majority of useful tasks that humans perform involve navigating to an object, grasping it, transporting it, and placing it somewhere else — often with the source and destination in different rooms. Mobile manipulation is the capability that enables this class of tasks.
The engineering challenge is that mobile manipulation is not simply "navigation + manipulation" — combining a mobile base with a manipulator arm creates coupling problems that each subsystem alone does not have. The arm's motion disturbs the base's stability. The base's motion creates errors in arm pose estimation. Grasping from a moving or recently-stopped platform introduces position error not present in fixed-base systems. Solving these coupling problems is the core technical challenge in mobile manipulation.
Platform Comparison
| Platform | Mobility | Arm | Price | Best For |
|---|---|---|---|---|
| Hello Robot Stretch 3 | Wheeled (omnidirectional) | Telescopic single 7-DOF | $25K | Healthcare, elderly assist, research |
| Boston Dynamics Spot + Arm | Legged (4-leg) | 6-DOF 4kg payload | $100K+ | Industrial inspection + manipulation |
| Mobile ALOHA (Stanford) | Wheeled (custom base) | Bimanual ALOHA arms | $32K (base+arms) | Bimanual mobile manipulation research |
| Unitree H1 + arm | Humanoid biped | Custom 7-DOF | $90K | Humanoid mobile manipulation research |
| Fetch Robotics FR2 | Wheeled (diff drive) | 7-DOF 6kg payload | $65K+ | Warehouse, logistics, shelf picking |
Whole-Body Control Approaches
Two dominant approaches exist for coordinating base and arm motion in mobile manipulation:
- MPC-Based Whole-Body Control: Treat the entire robot (base + arm) as a floating-base system and solve a joint trajectory optimization at 100–500Hz. This produces physically consistent base+arm motion that accounts for dynamic coupling. Computationally expensive: requires 20–100ms per solve on a modern processor for full humanoid systems (Unitree H1 + arm), which is tight for real-time control. Boston Dynamics and ETH Zurich's ANYmal implementations are the state of the art here.
- Hierarchical Control (virtual frame): Stabilize the base using a separate controller, then plan the arm trajectory in a "virtual frame" attached to the stabilized base. Computationally cheaper (two separate controllers rather than one joint optimization) and easier to tune. The tradeoff: doesn't fully account for dynamic coupling, so arm motions that disturb the base cause prediction errors. Preferred for wheeled systems where base dynamics are simpler.
Coordination Challenges
- Arm-Base Collision: The arm's workspace must be planned to avoid collision with the base, which requires a whole-body collision checker that includes both the arm and the base geometry simultaneously. This is often underimplemented — teams import standard arm collision checkers without updating them for the base geometry, resulting in collisions the planner didn't predict.
- Dynamic Stability: For legged systems, arm motion at high speed creates reaction torques that must be compensated by the legs. An arm moving at full speed with a 3kg payload creates 20–50Nm of reaction torque at the base — enough to destabilize a poorly-tuned legged system. Solutions: limit arm acceleration during locomotion, or couple the arm and leg controllers via whole-body MPC.
- Grasp from Moving or Recently-Stopped Base: Studies on Spot+Arm and similar platforms show that grasping accuracy is 3–5× worse when the base has stopped in the last 500ms, due to residual vibration and settling of the localization estimate. The practical fix: pause 1–2 seconds after stopping before initiating a grasp, or use force-torque feedback to compensate for position uncertainty during grasp approach.
Real Deployments
- Diligent Robotics Moxi (hospital logistics): Deployed at 10+ US hospitals including UT Southwestern, HCA Healthcare, and Ochsner. Moxi navigates hospital corridors and delivers supply trays to nursing stations — a navigation + transport task that doesn't require dexterous manipulation. This illustrates where mobile manipulation is commercially mature: transport and delivery rather than in-hand manipulation.
- 6 River Systems (warehouse): Chuck robots navigate alongside warehouse workers, carrying bins and reducing picker walking distance by 50–80%. Again: navigation + transport is the commercially deployed capability; arm manipulation is not yet deployed at this scale.
- Hello Robot (clinical trials, elderly assistance): Stretch robots are in multi-site clinical trials for in-home elderly assistance — reaching objects on shelves, opening doors, retrieving items. The simplicity of the Stretch design (single telescopic arm) is a feature here, not a limitation.
SVRC's solutions team can help scope mobile manipulation projects and connect you with the right platform for your application. See our solutions page for current offerings.