The field of robotics is witnessing significant advancements in simulation and control, with a focus on developing more realistic and generalizable models. Procedural generation of simulation assets is becoming increasingly important, enabling the creation of diverse and realistic environments for training and testing robots. Meanwhile, researchers are exploring new approaches to whole-body control, such as the use of skill libraries and latent spaces, to improve the ability of humanoids to interact with their environment. Reinforcement learning is also being applied to autonomous navigation, with the development of unified benchmarks and frameworks for training and evaluating navigation policies across diverse robotic platforms and operational environments. Furthermore, there is a growing interest in developing morphology-agnostic control policies, which can generalize to novel robotic embodiments and tasks. Notable papers in this area include:

• NavBench, which introduces a multi-domain benchmark for training and evaluating RL-based navigation policies,
• GCNT, which proposes a graph-based transformer policy for morphology-agnostic reinforcement learning.