Most robotics companies are trying to perfect manipulation and locomotion before they deploy. According to one robotics founder, this is a strategic error that’s costing them billions.
Nils, founder of Auki Labs, published an essay this week arguing that perception-first deployment is the winning strategy in robotics. The thesis is contrarian, coherent, and worth understanding.
The Three Capabilities
Robots can be categorized by three fundamental capabilities:
- Locomotion — Traversing environments (Tesla cars, delivery robots, elevators)
- Manipulation — Affecting objects (robotic arms, Weave laundry folders, assembly bots)
- Perception — Observing environments (CCTV cameras, inspection drones, crop monitors)
Most robotics attention and capital flows to manipulation and locomotion. The assumption: these are the “hard problems” that unlock the biggest value.
Nils argues this is backwards.
The Reliability Cliff
Here’s the core problem: manipulation and locomotion robots face a brutal “reliability cliff” that must be overcome before they can scale.
Consider an auto assembly line robot. When it malfunctions:
- The car might be damaged
- The entire assembly line might stop
- Downstream production is disrupted
- Costs compound rapidly
This means buyers demand extreme reliability — often “five 9s” (99.999% uptime) — before they’ll deploy at scale. Getting from 99% to 99.999% reliability is enormously expensive.
Self-driving cars face the same cliff. The technology might be 99% safe, but that 1% failure rate means people die. The last percentage points of reliability require massive additional investment.
Why Perception Is Different
Perception tasks have fundamentally different risk profiles:
- Lower cost of failure — A camera that misses a defect is bad, but not catastrophic
- Graceful degradation — Partial perception is still valuable
- Human backup viable — Flagged items can be reviewed by humans
- No physical harm — Perception robots rarely interact with the environment
This means perception robots can deploy earlier in their development cycle. They don’t need to overcome the reliability cliff before generating value.
The Budbreak Example
The essay highlights Budbreak, an agricultural robotics company focused on vineyards.
The naive assumption: Vineyard robots should harvest, prune, or spray (manipulation tasks).
The reality: Building a robot that harvests 10% better or cheaper than humans is extremely difficult. The manipulation requirements are complex, and the reliability cliff is steep.
The perception play: A single crop-destroying disease costs vineyards up to $60,000 per hectare annually. A rugged “Mars rover for vineyards” that detects disease early is:
- High value — $60K/hectare saved vs. marginal labor savings
- Low cost — Simpler technology than manipulation robots
- Low risk — If the rover misses something, humans can still inspect
There are tens of millions of addressable permanent crop acres in the US alone. The perception opportunity is massive.
The Strategic Framework
For any robotics deployment, analyze three dimensions:
| Dimension | Question |
|---|---|
| Value | Can the robot do the job cheaper or better? |
| Cost | What does development, deployment, and maintenance cost? |
| Risk | How catastrophic are failure modes? How much uptime is required? |
Perception tasks often score better on all three dimensions than commonly pursued manipulation tasks.
The Deployment Strategy
The perception-first approach suggests a roadmap:
- Deploy perception early — Lower reliability requirements mean faster market entry
- Capture territory — Establish customer relationships and domain expertise
- Collect data — Perception data trains manipulation models
- Add capabilities gradually — Layer manipulation on top of proven perception
This is how Tesla actually deployed their “robotics” — starting with perception (Autopilot), building reliability over time, and gradually expanding to more autonomous functions.
Investment Implications
If the perception-first thesis is correct:
Overvalued: Pure manipulation/locomotion plays burning capital on the reliability cliff
Undervalued: Perception-focused robotics with clear deployment paths and lower technical risk
Look for robotics companies that are generating revenue from perception tasks while building toward manipulation — not the reverse.
The Bottom Line
The contrarian case: most robotics companies are solving the wrong problem first. Perception tasks offer higher value, lower cost, and dramatically lower risk than manipulation tasks.
The winning strategy isn’t to perfect manipulation before deploying. It’s to deploy perception early, capture territory, collect data, and expand capabilities from a position of revenue and domain expertise.