Implementing Inverse Kinematics and a Walking System

1
ImplementingInverse Kinematics and a Walking
System

• Ideas presented by
• Rune Skovbo Johansen

2
Inverse Kinematics

• Local inverse kinematics control limbs such as
  legs and arms.
• Two or more bones per limb primary focus will
  be two bones, such as arms and legs in humanoids.
• IK in legs can be used to adapt walking
  animations to various terrain, speed and turning.
• IK in arms can be used to point at specific
  points, and to touch/grab specific objects.
• Primary focus on IK in legs.

3
Inverse Kinematics in Legs

• The inverse kinematics works locally on a limb,
  such as a leg, from hip to ankle, joined by the
  knee.
• Here, the hip is the root.
• Any bones extending from the IK limb (such feet
  or hands) could also be referred to as the
  trailing bones.
• The IK doesnt say anything about how the
  trailing bones move it is a problem in itself!

4
Adaptive Walking

• Adapting walking to the terrain and situation is
  more than just inverse kinematics
• Step positions (step markers) must be put on the
  ground, depending on path, speed, and terrain.
  They determine where the feet land.
• Anatomy of creature must be taken into account
• Number of legs
• Placement of legs on body
• Order in which the legs move (think spiders and
  bugs)
• The trajectory of the ankles must be interpolated
  between step markers.
• The alignment of the feet must be interpolated as
  well. (Movement of trailing bones is not
  determined by IK!)

5
Analysing terrain

• The terrain must be analysed to find suitable
  spots for step markers.
• This can be done using raycasts (support in
  engine?)
• ...or another method to easily examine and
  analyse geometry under the feet.
• Characters current velocity and rotational speed
  is also taken into account to predict a good step
  marker position.

6
Step Markers on Terrain
7
Ankle Trajectories

1. Global position of ankle (relative to ground) is
   recorded from keyframe data for whole walk-cycle.
2. Ankle trajectory is deduced from this.
3. The trajectories are stretched and bended
   according to placement of step markers.

8
Adapted Ankle Trajectories
9
Foot alignment

• Ankle trajectory is obtained from global position
  in key-frame data (relative to ground)
• Alignment of feet (or trailing bones in general)
  cannot be deduced this way.
• Local angles must be maintained (to some extent)
  from the key-frame data.

10
Foot alignment

• When on a step marker, a foot (trailing bones)
  must be transformed according to the step
  markers local space.
• When between step markers, a foot must be aligned
  according to local angles (from key-frame data).
• The system must know at which points the foot is
  touching the ground in the key-framed walk-cycle.
• The key-framed walk-cycle must be enriched with
  this data.

11

• Enriched walk-cycle with data (step-nodes) about
  at which points the feet are touching the ground.
• Must be provided by the animator.

12
The step-node starts having influence on the foot
alignment here, starting at 0.
13
The step-node has 100 influence on the foot
alignment here.
14
The step-node still has 100 influence on the
foot alignment here.
15
The step-node is back to 0 influence on the foot
at this point.
16
Hip Positions

• Since the positions of the feet are unpredictable
  when adapted, the positions of the hips must be
  adjusted according to the feet.
• Example The feet are so far from the hips that
  the legs are not long enough. The positions of
  the hips must be adjusted.
• Moving the hips just close enough is not good, as
  it gives a 100 stretched leg, which give poor
  movement in walk-cycles!
• The desired hip-ankle distance must be found in
  the key-framed data and applied. A solution that
  fits both legs as good as possible is preferred.
• Alternatively, local angles in hips and knees
  could be sought maintained, but this cannot be
  done without iterative approximation.

17
The Actual Inverse Kinematics

• When ankle-position and hip-position is found,
  the inverse kinematics are straightforward to
  apply for a two-bone limb, such as a leg.
• It can be solved exact.
• The knee points in same direction as it already
  did. 100 straight legs should be prohibited, but
  workarounds can be implemented anyway.

18
Walking Speed

• Characters that can walk with different speeds
  are more life-like.
• Not just walk or run.
• E.g. comes gradually to a stop, etc.
• Makes U-turns much more realictic.
• The system must support adjustments in speed
  without requiring separate walk-cycles for each
  speed.

19
Adjusting Walking Speed

• A creature can double its speed in two different
  ways
• Take twice as long steps(double step distance)
• Take steps twice as fast(double step frequency)
• What do humans do...?
• Answer A little of both!
• Sqrt(2) longer steps
• Sqrt(2) times as fast
• (Based on own unscientific tests involving
  walking a fixed distance at various speeds with a
  stop-watch while counting steps!)

20
Adjusting Walking Speed

• Engine already supports blending several
  walk-cycles this can be done according to
  characters current speed.
• Result of blending can be used as the basis of
  the adapted walking.
• The enriched data (step-nodes) must be taken into
  account in the blending though.

21
Additional Areas of Interest

• Inverse kinematics solutons for arms
• Extended automated walking system

22
IK Solutions for Arms

• IK in arms can be used to point at specific
  points, and to touch/grab specific objects.
• I havent thought as much about this.
• Touching/grabbing would often require that upper
  torso moves as well (bending down, reaching out)
  for humans.
• How could a generic flexible solution be
  implemented? Perhaps more conventional IK
  solutions.

23
Extended AutomatedWalking System

• Ability to analyse terrain and choose most
  suitable keyframed animation, based on step-node
  data in each.
• If animations walk straight, walk down, and
  walk up exist, most suitable is chosen (or
  blended between) based on terrain.
• Ability to treat hands as additional feet in some
  animations to enable e.g. climbing.

24
Engine Features I Need

• Query bone rotations in local space (at any
  point in time in animation).
• Query bone positions and alignment in model space
  (at any point in time in animation).
• Adjust bone rotations/alignments.
• Adjust speed of animations.
• Make raycasts or similar onto geometry.

25
Content I Need

• Several different models with standing, walking
  and running animations.
• Some models with skeletons with realistic human
  proportions.
• Some models that are not humanoid at all. (Pets,
  spiders, bugs, things with more than two legs.)

26
Questions or thoughts?