Hierarchical Planning & Contact-Aware Control Barrier Functions
Two Franka Emika Panda arms performing autonomous pick-up, contact-aware
handover, and smooth millimeter-scale placement in MuJoCo.
Right arm approaches with top-down orientation and establishes firm grasp
Contact-aware transfer with 10 mm distance sensing & synchronized clamping
Minimum-jerk interpolated carry to target with inertial-slip prevention
This thesis proposes a hierarchical collision-avoidance framework for dual-arm cable routing in simulation, handling three distinct safety aspects simultaneously.
Prevent the two 7-DoF manipulators from colliding with each other during coordinated motion.
Ensure neither robot arms nor the cable contact environmental obstacles (walls, fixtures, workspace boundaries).
Distinguish desired clip insertions from unintended collisions, enabling safe cable routing through fixtures.
A two-level planning and control pipeline: a global planner computes coarse collision-free paths, tracked by a local MPC controller wrapped in a safety-certified CBF.
DIT* or JIT in 14-DoF + DLO configuration space
MPC + Mode-Switching CBF-QP Safety Filter
Dual Franka Panda + Discrete Elastic Rods cable model
The CBF dynamically switches modes when force-torque sensors detect a
Contact Establishment Indicator (CEI) near routing clips to temporarily
relaxing hcable-env to allow fixture insertion
while maintaining all other safety guarantees.
Physics simulation via direct Python/C++ API without middleware overhead
Two 7-DoF manipulators from vikashplus/frankasim
Pinocchio-based differential inverse kinematics for real-time control
Discrete Elastic Rods for realistic cable bending, twisting & sagging
Control Barrier Functions via OSQP / qpOASES for certified safety
State-of-the-art asymptotically optimal global motion planners
Seamless hardware synchronization for future real-world deployment
MuJoCo dual-arm environment constructed via direct API. Rigid-body baseline with 19-phase pick-handover-place demo operational.
MPC tracking controller and core CBF constraints (arm-arm, environment, tension) using geometric distance queries.
DIT* or JIT adapted for DLOs with catenary-aware cost function and DLO-inclusive collision checking.
Force-simulated CEI implementation and mode-switching CBF logic for clip insertions.
Evaluation: fast motions, dynamic obstacle avoidance. Benchmarking against standard MPC and vanilla RRT.
Data analysis, final manuscript drafting, and defense presentation formatting.