Whole-Body Collision Avoidance for Articulated Robots

85 concepts60 papers1 proposal

When we think of a robot arm reaching for an object, our eyes naturally focus on the hand—the end-effector. Early robotic motion planners did the same, simplifying the robot to an infinitesimal point or a floating gripper. This abstraction works perfectly in wide-open spaces, but it fails catastrophically in the messy, crowded environments of the real world.

If you only track the tip of a whip, you cannot prevent the middle of the cord from striking a wall. Similarly, an articulated robot arm consists of bulky physical links and joints. As demonstrated by Jordan et al. and Murali et al., focusing exclusively on the end-effector's trajectory leaves the rest of the limb vulnerable. In cluttered workspaces, intermediate segments like the elbow can easily collide with obstacles, even if the hand reaches its destination perfectly. Whole-body collision avoidance shifts the planning paradigm from guiding a single point to safely navigating the robot's entire physical volume.

The Geometry of Configuration Space

How do we track and protect every point along a moving, twisting robotic arm without drowning in infinite geometric calculations? The foundational answer lies in transforming how we represent the world.

Instead of calculating collisions in the 3D physical space (the workspace), pioneering methods map the problem into the robot's internal state space: the Configuration Space (C-space). In C-space, every coordinate represents a complete set of joint angles. This mathematical flip provides a powerful simplification: the entire, multi-link robot collapses into a single, moving point.

Concurrently, physical obstacles in the room are inflated into high-dimensional "forbidden zones" or C-space obstacles. As explored by Brooks and Buckley, this mapping accounts for the non-linear relationship between joint rotations and physical positions.

Workspace (3D Physical Space)          Configuration Space (Joint Angles)
   [Obstacle]                             [C-Space Obstacle / Forbidden Zone]
       \                                         \
  (Robot Arm Links)  =======> Maps to =======>    * (Robot represented as a single point)
       /
   [Elbow]

By operating in this space, Grechanovsky et al. showed that a manipulator can safely "ride" along the edges of these forbidden zones. This ensures that intermediate joints remain clear of obstructions, bridging the gap between the Euclidean workspace and internal joint states. However, this transformation comes with a catch: mapping complex 3D environments into high-dimensional C-spaces is computationally expensive, and vital directional information needed for real-time avoidance can occasionally be lost, as noted by Klein et al..

Distributed Control and Whole-Body MPC

To achieve real-time safety without the heavy overhead of global C-space mapping, modern frameworks distribute the computational load across the robot's structure using reactive forces or dynamic optimization.

One intuitive mechanism is to treat the robot arm like a string of magnetic particles reacting to the environment. By distributing virtual control points across the entire kinematic chain rather than just the end-effector, each segment can independently experience repulsive steering forces from nearby obstacles. Becker et al. demonstrated that this distributed awareness allows sub-millisecond control loops to fluidly steer the entire robot body around unpredictable obstacles—such as a human coworker moving through a shared workspace—without abandoning the primary task.

For highly dynamic platforms, such as legged robots equipped with manipulator arms, safety requires balancing collision avoidance with physical equilibrium. In these scenarios, researchers utilize whole-body Model Predictive Control (MPC). By integrating signed distance fields or relaxed barrier functions directly into the optimization loop, frameworks designed by Chiu et al. and Gaertner et al. allow robots to discover collision-free motions on the fly. The robot can automatically duck its torso under a low ceiling or adjust a swinging leg to avoid tripping over low-lying obstacles while maintaining its balance.

The Enemy Within: Self-Collision Avoidance

Obstacles are not always external. For complex robotic systems, one of the most persistent hazards is the robot itself.

As early as 1973, Lewis et al. recognized that mounting a manipulator arm onto a mobile vehicle platform introduces severe self-collision risks; the vehicle's own wheels, motors, and structural components act as permanent obstacles that the arm must actively navigate around. When a system expands to dual-arm bimanual manipulation, the risk of self-interference and hardware damage increases exponentially, as highlighted by Grannen et al..

To manage this internal complexity, modern architectures unify external and internal obstacle avoidance under a single geometric framework. Divet demonstrated that a robot's own physical links can be modeled as dynamic virtual obstacles directly within the inverse kinematics loop, pruning self-intersecting configurations before commands are ever sent to the motors. Furthermore, by embedding the robot's physical topology into generative motion policies, as shown by Lv et al., bimanual systems can inherently respect spatial constraints between limbs, drastically reducing self-collisions during intricate coordination tasks.

The philosophy of whole-body safety ultimately extends beyond rigid metal links to encompass flexible continuum robots and the very objects being manipulated.

Continuum, or soft, robots do not have discrete joints; they bend and deform continuously along their entire structure. In delicate environments like minimally invasive surgery, focusing solely on where the tip of a catheter points is insufficient; the trailing backbone can easily snag or tear sensitive tissue, a problem highlighted by Chen et al.. To solve this, Ghoreishi et al. and Luo et al. utilized integral cost functions that evaluate the distance of the entire flexible structure from environmental boundaries, ensuring the soft robot safely snakes through tight spaces.

Rigid Whole-Body Avoidance             Soft/Continuum Whole-Body Avoidance
   O===O===O---> [Tip]                    (~~~~~~~~~~~~~~~~~~) ---> [Tip]
   ^   ^   ^                              ^    ^    ^    ^
[Discrete Control Points]               [Continuous Integral Distance Evaluation]

Finally, whole-body awareness must extend to coupled objects. When a robot manipulates a long, flexible cable or rope, the object effectively becomes an extension of the robot's own body. If the system only tracks the gripper, the middle of the cable remains prone to catching on surrounding objects.

To prevent this, Yu et al. developed frameworks that apply repulsive forces to both the robot's kinematic chain and discrete points along the manipulated object. This holistic approach ensures that the robot, the tool, and the payload act as a single, coordinated, collision-aware system. Similarly, Pallar et al. demonstrated that modeling secondary components—such as the swinging cables connecting multi-drone transport teams—as explicit geometric constraints prevents catastrophic entanglements in tight spaces.

Key Foundational Work

Planning Considerations for a Roving Robot with Arm (1973) — contributed 3 concept elements to this direction.
An Algorithm for Moving a Computer-Controlled Manipulator While Avoiding Obstacles (1983) — contributed 2 concept elements to this direction.
Find-Path for a PUMA-Class Robot (1983) — contributed 1 concept elements to this direction.
History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 59 papers
  2. Version 1 · 2026-10-04 · 89 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_full_body_collision_avoidance_9f306637,
  title = {full-body collision avoidance},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/full-body-collision-avoidance/rdr_9f306637fadf41c691b182ea364b5db8},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 2, frozen text):

@misc{lacuna_full_body_collision_avoidance_9f306637_v2,
  title = {full-body collision avoidance},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/full-body-collision-avoidance/rdr_9f306637fadf41c691b182ea364b5db8/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake