Robotics Simulation
Overview
Robotics simulation studies how to represent robot bodies, sensors, environments, and interactions so researchers can test control and learning safely and repeatably before hardware deployment. Approaches range from rigid-body and contact physics engines such as MuJoCo and Isaac Sim, to finite-element models for soft materials, differentiable or learned dynamics, and generated scenes for navigation and manipulation. GPU-parallel reinforcement learning, synthetic demonstrations, digital twins, and simulation-in-the-loop planning help scale training, while benchmarks and hardware-in-the-loop tests assess transfer. The central tension is balancing physical and sensor fidelity with speed and reproducibility: visual realism alone does not ensure correct interactions, and simulated success does not always predict real-world performance without calibration and validation.
Latest papers
351 papers34 papers in the last four weeks, down 45% on the four weeks before. 0.3% of all new papers.
CableRobotGraphSim: A Graph Neural Network for Modeling Partially Observable Cable-Driven Robot Dynamics
CableRobotGraphSim, a novel Graph Neural Network (GNN) model for cable-driven robots that aims to address shortcomings of prior simulation solutions. By representing cable-driven robots as graphs, with the rigid-bodies as nodes and the cables and contacts as edges, this model can quickly and accurately match the properties of other simulation models and real robots, while ingesting only partially observable inputs. Accompanying the GNN model is a sim-and-real co-training procedure that promotes generalization and robustness to noisy real data. This model is further integrated with a Model Predictive Path Integral (MPPI) controller for closed-loop navigation, which showcases the model's speed and accuracy.