Backflips are easy, stairs are hard: Robots still struggle with simple human movements, experts say

At Fortune’s Brainstorm AI conference, Sequoia partner Stephanie Zhan and Skild AI CEO Deepak Pathak cautioned that flashy humanoid-robot demos overstate real-world capability—robots perform well on complex tasks in controlled environments but still fail at seemingly simple, interactive tasks like picking up a glass or climbing stairs. They said the industry is shifting from hand‑programmed machines to learning-based, more general-intelligence platforms that could be re-tasked across factory automation, consumer services and hazardous work, potentially expanding addressable markets and helping alleviate U.S. blue‑collar labor shortages and reshoring challenges. However, speakers flagged social and deployment risks—job displacement and the uncertain timing and economics of truly general-purpose robots—which are critical for investors weighing bets on adaptable robotics platforms versus specialized automation.

Analysis

At Fortune’s Brainstorm AI conference, Sequoia partner Stephanie Zhan and Skild AI CEO Deepak Pathak cautioned that viral humanoid-robot demos overstate real-world capability: backflips and sidewalk skipping look impressive, but robots still fail at interactive tasks such as picking up a bottle or climbing stairs. Zhan paraphrased Hans Moravec—"what looks hard is easy, but what looks easy is really hard"—and Pathak emphasized that many demos operate in controlled environments where the robot is not interacting with objects.

Speakers described a structural shift from hand-programmed robots to systems that learn from experience, which could unlock adaptable platforms deployable across factories, consumer households and hazardous tasks; Qualcomm CEO Rene Haas specifically noted humanoid adaptability could outcompete fixed robotic arms in some factory roles. The thematic signals show mildly positive market sentiment (sentiment_score 0.25) but a modest market impact (market_impact_score 0.35), reflecting optimism tempered by execution risk.

Key investor risks are timing and economics of true general-purpose robots, gaps in vision and continuous interaction needed for manipulation, and social/regulatory fallout from job displacement. Investors should therefore demand real-world validation in uncontrolled environments and prioritize software, perception, and transfer-learning capabilities over flashy single-task hardware demos.

Key Decisions for Investors

• Prioritize exposure to companies demonstrating learning-based control, robust vision/perception stacks and repeatable validation in unstructured, real-world tests, rather than firms relying mainly on staged demos
• Favor platform- and software-led robotics players and diversified industrial suppliers that enable re-tasking (simulation, transfer learning, control software) over single-purpose consumer or humanoid hardware pure-plays
• Defer large allocations to pure-play humanoid hardware until commercial deployments prove reliable object manipulation and stair/terrain negotiation in uncontrolled settings
• Monitor labor-market, reshoring and regulatory signals as downside catalysts and keep position sizes modest or hedged given the modest market-impact signal and uncertain timing