- Published on
The Dexterity Debate: Roboticists Clash Over the Future of Humanoid Hands-On Work
- Authors
- Name
- Humanoids daily
- @humanoidsdaily
A provocative new essay from Rodney Brooks, a foundational figure in robotics known for co-founding iRobot, has ignited a sharp debate over the future of humanoid machines. In "Why Today’s Humanoids Won’t Learn Dexterity," Brooks argues that the current, vision-centric approach to training robots is fundamentally flawed, and that without a rich sense of touch, true human-like manipulation is an unattainable fantasy. This skeptical take has drawn pointed responses from experts who believe utility will precede perfection.
The Case for Touch
Brooks’ core argument centers on what he sees as a critical oversight in the industry's race to build general-purpose humanoids. Companies like Tesla and Figure, he notes, are heavily invested in training robots by feeding them video data of humans performing tasks. While this approach has produced visually impressive demonstrations, Brooks contends it misses the most crucial ingredient for dexterity: haptic feedback.
To illustrate his point, he highlights a stark experiment where a person adept at lighting a match with full sensation struggles profoundly after their fingertips are anesthetized. The task, once effortless, becomes a clumsy, frustrating ordeal. This, Brooks argues, is the state of today's humanoids—they are operating without the rich, complex data stream that comes from the 17,000 mechanoreceptors in a human hand. He posits that the "end-to-end learning" models that worked for language and vision won't translate to physical tasks because the "ends"—the sensory inputs—are incomplete. Without the ability to feel pressure, slip, and texture, robots will remain clumsy mimics, unable to handle the unpredictability of the real world.
The "Low-Hanging Fruit" Counter-Argument
Responding to Brooks' essay, robotics expert Dr. Scott Walter characterized the critique as a "High-Hanging Fruit Argument." In a post on X, Walter argued that Brooks is creating a false dichotomy: "Since humanoids can't do everything we can do, they can't do anything we can do."
He is using the High-Hanging Fruit Argument: Since humanoids can't do everything we can do, they can't do anything we can do It's a bad faith argument He describes a lot of tasks too difficult to do without touch sensing But for every example, I can find a counter-example🧵
Robotics expert Rodney Brooks doesn't believe that today's humanoid robots will ever learn human-like dexterity... He thinks @Tesla and @Figure_robot are wrong. What do you think, Scott? @GoingBallistic5. His arguments (see his blog article): rodneybrooks.com/why-todays-hum…
Walter suggests this is a "bad faith argument." He points out that for every complex task requiring fine-tuned touch sensing, there are countless simpler, valuable tasks that don't. "Many workers wear protective gloves which dull the senses. Yet they still manage to perform," he wrote. The key, according to Walter, is not perfect human replication, but utility.
He draws a parallel to one of Brooks' own inventions: the Roomba. "We can all agree the Roomba, does at best 80% of the vacuuming, is much slower, and less performative than humans," Walter notes. "So why are people still buying them? Because they provide just enough utility to be useful." He argues humanoids will follow the same path, tackling the "low-hanging fruit" first and improving over time as technology advances.
A Tale of Two Futures
The debate encapsulates two starkly different visions for the near future of robotics. On one side is Brooks' pragmatic skepticism, which foresees the current multi-billion dollar bet on vision-only humanoids leading to a dead end. He predicts the "humanoid" robots of the next 15 years will look very different, likely incorporating wheels and specialized grippers rather than mimicking the human form so closely.
On the other side is a vision of exponential progress, articulated in a recent, widely-shared article by Adam Dorr of RethinkX. In "Living Like Kings," Dorr paints a picture of a world transformed by artificial labor. He argues that by the 2030s, humanoid robots will not be single-task machines but generalist experts—a "five-star chef, olympic-coach-calibre fitness trainer, master carpenter, [and] bonded electrician" all in one. This abundance of skilled, virtually free labor, Dorr writes, will "democratize luxury" and create a world where everyone enjoys a standard of living reserved for royalty throughout history.
This utopian outlook is precisely the kind of hype that Brooks' essay seeks to temper. While Dorr's vision relies on the rapid, successful scaling of general-purpose humanoids, Brooks' analysis suggests the foundational technology to make that leap doesn't yet exist.
The resolution of this debate will define the next decade of automation. Will the industry find, as Walter suggests, that "useful work is all that counts" and that today's humanoids are good enough to get started? Or will the stubborn complexities of physical interaction prove Brooks right, forcing a fundamental rethink of what a general-purpose robot should be? For now, the question of whether humanoids can truly get a grip on the physical world remains firmly in hand.