"Latent-Space Prediction Error and Fuzzy Risk-Scaled Recovery for Robotic Manipulation," by Thanh-Hai Tran and Dr. Xuan-Bach Le, has been accepted at ACIVS 2026 in Okinawa, Japan, December 2026. Industrial pick-and-place lines tolerate almost no failures, yet most safety layers today only fire after a fault has already surfaced — and they respond with an emergency stop that makes no distinction between a minor wobble and a serious fault. The authors ask whether latent prediction error can instead serve as a graded recovery signal. The closed loop pairs a frozen DINOv2 encoder with an RSSM whose KL divergence supplies an online risk reading, an ANFIS risk mapper, and a controller that applies bounded trajectory corrections scaled by that risk coefficient. Evaluated in Webots with a UR5e arm, the gains concentrate on the harder disturbances: Jerky (+25.0 percentage points) and Speed (+42.5 percentage points), with mean joint deviation falling 15.4–44.8% under Vibration, Jerky and Pause. CPU-only latency reaches P95 = 34.24 ms, 65.8% under the 100 ms design budget. Paper: https://lexuanbach.github.io/publication/ACIVS2026a.pdf Supplement: https://lexuanbach.github.io/publication/ACIVS2026a-supplement.pdf Code: https://github.com/haitran-13/acivs2026-supplementary