Revision history [back]

eband_local_planner (Classical Elastic Band approach by Quinlan et al.) and teb_local_planner (Timed Elastic Band (TEB) approach) are two completely different planning algorithms. However, the TEB principle is based on the classic elastic band idea.

eband_local_planner (Elastic Band)

• Local path deformation (path: no timing law) based on internal and external forces
• Internal forces contract the path (-> leading to the shortest path between start and goal)
• External forces repel the path from obstacles
• Implementation based on bubbles that represent discrete path points and free-space.
• Adaption of the trajectory length w.r.t. bubbles/free-space (insertion and deletion of discrete points)
• Extension to non-holonomic kinematics (supports differential-drive and omnidirectional robots)
• Subject to local minima (e.g. left or right path around an obstacle, depends on initial path)

teb_local_planner (Timed Elastic Band)

• Local trajectory deformation/optimization (trajectory: includes temporal information)
• Instead of generating and applying forces, an objective/cost function is minimized
• Temporal information is subject to optimization -> time-optimal trajectories (replacement for the internal forces)
• Temporal information allows incorporation of (kino-)dynamic constraints during optimization (no need for a dedicated path-following controller, the teb_local_planner mimics a predictive controller)
• Adaptation of the trajectory length based on the temporal discretization (insertion and deletion of discrete trajectory points)
• Supports differential-drive, car-like and omnidirectional robots
• Explores multiple distinctive topologies for parallel trajectory optimization in order to partially overcome the local minima problem (only in the scope of the local costmap due to limited CPU resources, a global planner is still required).
• Path-following mode (minimize distance to global plan instead of minimizing transition time)
• Bottleneck: very high computational burden (-> limited map size/resolution resp. robot size).

eband_local_planner (Classical (classical Elastic Band approach by Quinlan et al.) and teb_local_planner (Timed Elastic Band (TEB) approach) are two completely different planning algorithms. However, the TEB principle is based on the classic elastic band idea.

eband_local_planner (Elastic Band)

• Local path deformation (path: no timing law) based on internal and external forces
• Internal forces contract the path (-> leading to the shortest path between start and goal)
• External forces repel the path from obstacles
• Implementation based on bubbles that represent discrete path points and free-space.
• Adaption of the trajectory length w.r.t. bubbles/free-space (insertion and deletion of discrete points)
• Extension to non-holonomic kinematics (supports differential-drive and omnidirectional robots)
• Subject to local minima (e.g. left or right path around an obstacle, depends on initial path)

teb_local_planner (Timed Elastic Band)

• Local trajectory deformation/optimization (trajectory: includes temporal information)
• Instead of generating and applying forces, an objective/cost function is minimized
• Temporal information is subject to optimization -> time-optimal trajectories (replacement for the internal forces)
• Temporal information allows incorporation of (kino-)dynamic constraints during optimization (no need for a dedicated path-following controller, the teb_local_planner mimics a predictive controller)
• Adaptation of the trajectory length based on the temporal discretization (insertion and deletion of discrete trajectory points)
• Supports differential-drive, car-like and omnidirectional robots
• Explores multiple distinctive topologies for parallel trajectory optimization in order to partially overcome the local minima problem (only in the scope of the local costmap due to limited CPU resources, a global planner is still required).
• Path-following mode (minimize distance to global plan instead of minimizing transition time)
• Bottleneck: very high computational burden (-> limited map size/resolution resp. robot size).

eband_local_planner (classical Elastic Band approach by Quinlan et al.) and teb_local_planner (Timed Elastic Band (TEB) approach) are two completely different planning algorithms. However, the TEB principle is based on the classic elastic band idea.

eband_local_planner (Elastic Band)

• Local path deformation (path: no timing law) based on internal and external forces
• Internal forces contract the path (-> leading to the shortest path between start and goal)
• External forces repel the path from obstacles
• Implementation based on bubbles that represent discrete path points and free-space.
• Adaption of the trajectory length w.r.t. bubbles/free-space (insertion and deletion of discrete points)
• Extension to non-holonomic kinematics (supports differential-drive and omnidirectional robots)
• Subject to local minima (e.g. left or right path around an obstacle, depends on initial path)

teb_local_planner (Timed Elastic Band)

• Local trajectory deformation/optimization (trajectory: includes temporal information)
• Instead of generating and applying forces, an objective/cost function is minimized
• Temporal information is subject to optimization -> time-optimal trajectories (replacement for the internal forces)
• Temporal information allows incorporation of (kino-)dynamic constraints during optimization (no need for a dedicated path-following controller, the teb_local_planner mimics a predictive controller)
• Adaptation of the trajectory length based on the temporal discretization (insertion and deletion of discrete trajectory points)
• Supports differential-drive, car-like and omnidirectional robots
• Explores multiple distinctive topologies for parallel trajectory optimization in order to partially overcome the local minima problem (only in the scope of the local costmap due to limited CPU resources, a global planner is still required).
• Path-following mode (minimize distance to global plan instead of minimizing transition time)
• Bottleneck: very high computational burden (-> limited map local costmap size/resolution resp. robot size).