Discovering and synthesizing humanoid climbing movements

Abstract

Fig. 1. Two paths (green and blue arrows) to the top hold in a bouldering problem, as discovered by our method. Compared to previous work, we are not limited to only moving one limb at a time or having the climber's hands and feet on predefined climbing holds; limbs can also hang free for balance, or use the wall for friction. The dashed rectangles highlight the one-to-many mapping from stances (assignments of holds to limbs) to climber states. This paper addresses the problem of offline path and movement planning for wall climbing humanoid agents. We focus on simulating bouldering, i.e. climbing short routes with diverse moves, although we also demonstrate our system on a longer wall. Our approach combines a graph-based highlevel path planner with low-level sampling-based optimization of climbing moves. Although the planning problem is complex, our system produces plausible solutions to bouldering problems (short climbing routes) in less than a minute. We further utilize a k-shortest paths approach, which enables the system to discover alternative paths-in climbing, alternative strategies often exist, and what might be optimal for one climber could be impossible for others due to individual differences in strength, flexibility, and reach. We envision our system could be used, e.g. in learning a climbing strategy, or as a test and evaluation tool for climbing route designers. To the best of our knowledge, this is the first paper to solve and simulate rich humanoid wall climbing, where more than one limb can move at the same time, and limbs can also hang free for balance or use wall friction in addition to predefined holds.

Figure Skating Simulation from Video

“…Al Borno et al [ABDLH13] successfully generated many motions, including walking and spinning, using simple objectives according to a window optimized using co‐variance matrix adaptation evolution strategy (CMA‐ES). Climbing has been simulated using CMA‐ES without the use of motion capture data [NRH17]. Swimming has been reproduced using a CPG‐based locomotion controller that can generate muscle contraction signals automatically [SLST14].…”

Learning Physically Based Humanoid Climbing Movements

“…One interesting type of this problem is motion planning for climbing agents. It has various applications including control of rescue robots [Bre06a] and procedural generation of animations in video games [NRH17, LMS09a]. In this context, the main goal is to produce physically‐feasible and natural climbing movements while maintaining the balance of the character.…”

“…Top: Solution involving a dynamic leap (dashed circles) emerges when we allow moving all 4 limbs at the same time. Middle: A different strategy where we restrict movement to only 2 limbs at a time, similar to the previous work of [NRH17]. Bottom: The system of [NRH17] produces more cumbersome movements, and in this problem does not find a solution until after 10 minutes of failed attempts.…”

“…Middle: A different strategy where we restrict movement to only 2 limbs at a time, similar to the previous work of [NRH17]. Bottom: The system of [NRH17] produces more cumbersome movements, and in this problem does not find a solution until after 10 minutes of failed attempts. The top and middle solutions were found at the first try (less than 1 minute of CPU time), due to our neural network components trained through random simulation practice.…”

“…Our work builds on recent state‐of‐the‐art work in [NRH17], which combines a high‐level graph‐based planning of movement sequences and a low‐level sampling‐based controller for synthesizing individual movements. While that system is able to solve bouldering problems and synthesize plausible climbing movements, it has three main limitations: 1) the high‐level path planner uses several handcrafted heuristics to cull out transitions that might be unnatural or impossible, which is inelegant and raises questions about generalizability of the approach, 2) the sampling‐based low‐level controller does not employ any information about past movements, which makes it prone to failures in challenging moves, and 3) the system is only demonstrated in movements where at most two limbs move to new holds at the same time.…”