Wireless connectivity frees factory machinery from fixed cabling, but electromagnetic interference causes packet loss and delay, which makes real-time remote control of robots from the Edge or Cloud difficult. The paper introduces FoReCo, a recovery mechanism that uses machine learning forecasting to infer the control commands lost to interference. FoReCo is evaluated in simulation and experimentally over interference-prone IEEE 802.11 links with a commercial research robot performing pick-and-place tasks. Under interference it reduces the trajectory error by up to 18 times in simulation and 2 times in experiments, and it is lightweight enough to run on hardware already used in existing solutions.