Wheeled Mobile Manipulators: Why Robots Like Dex Win

Grand View Research pegs the current humanoid robot market at $1.6 billion, and estimates wheeled humanoid-like robots make up 65% of it. That split tells you something practical: the robots getting real work done right now often roll, not walk.

Richtech Robotics’ Dex is a good example of where the industry is heading. It’s a wheeled autonomous mobile robot (AMR) with two arms, designed for commercial and industrial environments where uptime, safety, and repeatable performance matter more than robot aesthetics. If you’re leading operations, automation, or innovation, this is the category to watch—because it’s where AI and robotics are already transforming industries worldwide.

What follows is the business case for wheeled mobile manipulators, what Dex signals about the next phase of physical AI, and a grounded checklist for deciding whether this class of robots belongs in your facility in 2026.

Wheeled mobile manipulators are winning because reliability beats novelty

Answer first: Wheeled mobile manipulators are gaining market share because they deliver the core promise of robotics—repeatable work with predictable uptime—at a lower energy and deployment cost than legged systems.

Humanoid robots are impressive engineering feats. They also bundle a long list of failure points: balance control, high-torque joints, complex gait planning, more actuators, more wiring, and more maintenance. That complexity isn’t “bad”; it’s just expensive and fragile when you’re trying to run a business.

Wheels simplify the hardest part of biped robots: staying upright. In practical terms, that simplification cascades into:

• Longer usable runtime (less energy burned on balance and locomotion)
• Faster iteration (fewer mechanical edge cases to debug)
• Higher payload stability (less dynamic sway, fewer drops)
• Lower operational risk in human-populated spaces

Richtech’s president, Matt Casella, framed Dex’s design choice in a way most operators will appreciate: battery life and operational reliability drive adoption. The company claims Dex can operate 4+ hours continuously on its wheeled platform, while maintaining stability and moving safely in tight spaces.

My take: the first wave of “humanoid deployments” that scale broadly won’t look like sci-fi bipeds. They’ll look like AMRs with arms—because that’s the shortest path from demo to daily use.

Dex is a signal: mobile manipulation is becoming a product category, not a project

Answer first: Dex shows that mobile manipulation is shifting from custom integration work to a repeatable product pattern: AMR base + dual-arm manipulation + on-board AI compute + simulation-driven deployment.

Dex sits in a growing middle ground:

• More capable than a delivery AMR (it can manipulate objects)
• Less complex than a walking humanoid (it doesn’t need legs)

That middle ground matters because most businesses don’t need a robot that can climb stairs. They need a robot that can move through the facility, pick/place reliably, interact safely with people, and recover gracefully when something changes.

The “two arms on an AMR” advantage is real

A single arm can do a lot. Two arms change the task list.

Two arms enable:

• Bimanual handling (holding a box with one hand while cutting tape with the other)
• Stabilization (one arm braces while the other manipulates)
• Faster cycle completion for multi-step service tasks (e.g., sorting + placing)

This is also why wheeled humanoid-like robots show up in service settings: they’re not trying to be human. They’re trying to complete a workflow without handing off steps to a person.

Why the telescoping body matters more than it sounds

Dex’s design includes a telescoping “neck,” which is a quiet but important design decision. Fixed-height robots force you to redesign work cells. Adjustable-height robots adapt to the environment.

If you want adoption beyond pilot projects, robots have to fit into existing spaces—not demand a remodel.

AI inside the robot is the difference between “moves” and “works”

Answer first: The AI stack (vision, navigation, language, and policy control) is what turns a mobile manipulator from a moving platform into an autonomous worker.

Richtech’s Dex architecture, as described publicly, is built around on-device compute and a modern autonomy toolchain:

• NVIDIA Jetson Thor for on-board AI processing
• Lidar-based SLAM for navigation and mapping
• Real-time obstacle detection (described at millisecond scale)
• Natural language processing for customer or worker interaction
• Edge decision-making without constant cloud connectivity
• Simulation workflows (via NVIDIA Isaac Sim) to train and test before deployment

Here’s what that adds up to operationally: fewer brittle “if this, then that” scripts—and more behavior that generalizes when the world isn’t perfectly staged.

The real ROI isn’t the robot—it’s the recovery behavior

Most automation ROI calculations focus on the happy path: cycle time, throughput, labor savings.

In practice, the biggest differentiator is how often a system:

• Gets confused by clutter
• Fails to grasp odd objects
• Blocks aisles
• Needs human rescue
• Requires a technician visit

Physical AI earns its keep when the robot can recognize it’s stuck, choose a safe fallback, and ask for help in a structured way. That’s why edge compute, strong perception, and simulation-based testing matter.

If your robotics vendor can’t describe their recovery strategy in plain language, you’re buying a demo.

Where wheeled mobile manipulators fit best in 2026

Answer first: The strongest near-term fits are environments with repeatable movement, moderate manipulation complexity, and chronic labor gaps—especially where humans and robots share space.

The original reporting highlights service-industry labor shortages and Richtech’s early validation with food delivery robots (with 80+ customers in that early phase). That history matters because it suggests a playbook: start where the environment is semi-structured, prove reliability, then move into more demanding industrial tasks.

Here are the most realistic use cases I see for wheeled mobile manipulators over the next 12–24 months.

Manufacturing: internal logistics plus light work-cell tasks

Think “move + do,” not “replace the whole line.” Examples:

• Deliver totes to a station and present parts at the right height
• Remove finished items and place them in bins
• Perform simple kitting (when parts are standardized)

Best fit: facilities where walking time and task handoffs are killing productivity.

Warehousing: exception handling and assisted picking

Pure AMRs move shelves and pallets well. The gap is the messy middle: exceptions, mis-sorts, damaged packaging.

A mobile manipulator can:

• Clear jams or handle exceptions with human oversight
• Perform light rework or repack tasks
• Support pick/pack in constrained zones

Best fit: operations with high SKU variability but defined storage standards.

Hospitality and retail: “staff multiplier” tasks

Richtech has lived in this world already. A wheeled mobile manipulator makes sense when it reduces constant micro-interruptions:

• Restocking lightweight items
• Running items to staff areas
• Basic customer-facing interactions where speech is simple and bounded

Best fit: high-foot-traffic environments where safety, smooth navigation, and polite interaction matter.

RaaS now, DaaS next: the business model shift operators should plan for

Answer first: Robot-as-a-Service (RaaS) reduces adoption friction today, but Data-as-a-Service (DaaS) will shape vendor power tomorrow—because physical AI improves with real-world operational data.

Richtech’s model centers on RaaS deployments, with direct sales available when it suits the customer. That’s increasingly common because it aligns incentives: the vendor gets paid when the robot stays deployed and productive.

But the more strategic point is Casella’s long-term intent: evolve toward data as a service. This is where many buyers need to slow down and get specific.

Why robotics data becomes a competitive moat

Physical AI systems need real operational data: sensor streams, environment variability, interaction logs, near-misses, failure cases. The company that runs more robots in more environments learns faster.

If a vendor has 1,000+ deployed robots, they can:

• Improve navigation robustness
• Improve grasp success rates
• Expand task libraries faster
• Reduce edge-case failures across the fleet

That’s good for you—if you benefit from those improvements.

The buyer-side risk: data rights and lock-in

If you’re deploying mobile manipulators in 2026, you should negotiate and document:

• Who owns the operational data (raw and derived)
• Whether your data is used to train models for other customers
• Data retention periods
• Security posture and access controls n- Export options if you switch vendors

I’m firmly pro-RaaS for many organizations, but I’m also pro-clarity: your robotics contract is becoming a data contract.

A practical adoption checklist for wheeled mobile manipulators

Answer first: Successful deployments start with task selection, facility readiness, and safety + support planning—not with a robot spec sheet.

If you’re evaluating a platform like Dex (or any wheeled mobile manipulator), use this as your first-pass filter.

1) Choose tasks that survive variability

Good starter tasks:

• Repetitive routes with stable floor conditions
• Objects with consistent geometry (bins, trays, standardized cartons)
• Clear success criteria (delivered, placed, returned)

Avoid as a first deployment:

• Highly reflective, cluttered spaces with constant layout changes
• Unbounded manipulation (miscellaneous items in a heap)
• Tasks requiring fine force control without strong fixturing

2) Demand proof of recovery, not just autonomy

Ask vendors to demonstrate:

• What happens when an aisle is blocked
• What happens when a grasp fails twice
• How the robot requests help
• How a supervisor clears errors

3) Plan for humans first

These robots succeed when they reduce worker strain and interruptions. Set expectations early:

• Define “robot zones” and right-of-way rules
• Train staff on safe interaction
• Create a simple escalation path when the robot needs help

4) Measure the right KPIs

Early KPIs that predict scale:

• Interventions per shift
• Mean time to recovery
• Successful task completion rate
• Uptime by day and by location
• Worker satisfaction in impacted areas

Cost savings matter, but reliability metrics tell you if you’re headed toward expansion—or stuck in pilot purgatory.

What Dex means for the broader “AI & Robotics” industry story

Wheeled mobile manipulators are a clean illustration of the theme running through this series—Artificial Intelligence & Robotics: Transforming Industries Worldwide. The transformation isn’t about a single flashy robot. It’s about systems that combine mobility, manipulation, and AI into a dependable workflow tool.

Dex also points to where competition will intensify next: not just who can build the robot, but who can deliver repeatable deployments, strong support, and compounding improvements from fleet learning. The companies that win won’t be the ones with the fanciest demo. They’ll be the ones whose robots show up every day and finish the shift.

If you’re planning automation investments for 2026, here’s the question to pressure-test your roadmap: Which workflows in your operation are stable enough for a wheeled mobile manipulator today—and costly enough in labor and interruptions that you’ll feel the impact within one quarter?