Real-Time Fully On-Board State Estimation for Six-Bar Tensegrity Robots

State estimation for six-bar tensegrity robots remains challenging even with external vision systems, due to self-occlusion. When restricted to onboard sensing only, the problem becomes significantly harder, as the neglect of axial rotation of bars in modeling, state discontinuities from intermittent contacts, and structural symmetry all introduce solution non-uniqueness. We propose a two-layer state estimation framework to address these challenges. The first layer employs a Contact-Aided Invariant Extended Kalman Filter (CA-InEKF) to estimate the pose of the robot's body frame with respect to the world frame. To resolve ambiguities arising from structural symmetry and the neglect of axial rotation of bars, the second layer estimates the positions of all nodes within the robot's body frame by enforcing geometric constraints and applying a dual-bar joint correction mechanism, which by leveraging the poses of two connected bars in the world frame resolves axial rotation ambiguity and breaks structural symmetry. Simulation results on the Isaac Sim platform validate the effectiveness of the proposed algorithm: even after drop impacts, the system rapidly recovers and maintains a steady-state position RMSE below 0.1 m.