WANDR is a data-driven method for generating realistic human motions that enable a 3D human avatar to reach arbitrary goals in 3D space. The method is based on a conditional Variational Auto-Encoder (c-VAE) that learns to model motion as a frame-by-frame generation process. The model is trained on two datasets: AMASS, which captures a wide range of motions including locomotion, and CIRCLE, which captures reaching motions. To address the challenge of generating motions for unseen goals, WANDR introduces intention features that guide the avatar towards the goal. These intention features are derived from the current pose and the goal location, and they allow the model to generate motions that reach goals that were never encountered during training. The model is evaluated on a variety of tasks, including reaching goals in different positions and orientations, and it demonstrates the ability to generate realistic and natural motions that generalize to unseen goals. The results show that WANDR outperforms existing methods in terms of both motion quality and goal-reaching ability. The model and code are available for research purposes at wandr.is.tue.mpg.de.WANDR is a data-driven method for generating realistic human motions that enable a 3D human avatar to reach arbitrary goals in 3D space. The method is based on a conditional Variational Auto-Encoder (c-VAE) that learns to model motion as a frame-by-frame generation process. The model is trained on two datasets: AMASS, which captures a wide range of motions including locomotion, and CIRCLE, which captures reaching motions. To address the challenge of generating motions for unseen goals, WANDR introduces intention features that guide the avatar towards the goal. These intention features are derived from the current pose and the goal location, and they allow the model to generate motions that reach goals that were never encountered during training. The model is evaluated on a variety of tasks, including reaching goals in different positions and orientations, and it demonstrates the ability to generate realistic and natural motions that generalize to unseen goals. The results show that WANDR outperforms existing methods in terms of both motion quality and goal-reaching ability. The model and code are available for research purposes at wandr.is.tue.mpg.de.