What happened: IEEE Spectrum’s robotics Video Friday rounded up fresh clips and research notes, including “LATENT,” a system that learns humanoid tennis skills from imperfect human motion data instead of pretending every rally comes with perfect mocap.

Why it matters: Athletic, contact-rich skills are exactly where humanoids go from “cool demo” to “oops, the ball hit my face.” The piece is a reminder that action data is still the bottleneck, and the workaround is usually some form of messy human demonstration plus learning.

Wider context: The roundup also highlights the industry’s current coping strategy: shared autonomy. Humans issue higher-level triggers, robots execute gnarly finger-and-force choreography, and everyone calls it “scalable” while quietly collecting training data.

Background: Video Friday is Spectrum’s weekly robotics digest, mixing research previews, lab demos, and occasional corporate flexing into one feed—useful precisely because it shows what’s real (videos) and what’s still vapor (marketing adjectives).

Videos: Tennis Playing Humanoid Robot, Horse Quadruped — IEEE Spectrum

Droid Brief Take: Humanoids don’t need more inspirational “general-purpose” speeches—they need more boring, unglamorous action data, preferably gathered before the robot yeets a tennis ball into your orbital socket. Resistance is futile, but bruises are optional.

Key Takeaways:

• LATENT’s premise: The system targets tennis on humanoids, explicitly calling out that perfect humanoid action data and clean human kinematic data are hard to get in tennis scenarios, so it learns from imperfect motion data instead.
• Shared autonomy is everywhere: The roundup describes approaches where humans don’t teleoperate every finger; they trigger pre-learned primitives (via keyboard/pedal-type inputs) so robots can execute contact-rich manipulation while training data collection stays feasible.
• Reality in video form: As a weekly video digest, the post is valuable because it’s not just claims—it’s demonstrations, paper links, and lab context, which makes it easier to separate “working prototype” from “fundraising mood board.”